SYSTEM FOR DETERMINING THE STATUS OF A GAS CYLINDER

Abstract

According to the invention there is provided a system for determining a status of a gas cylinder, the system comprising: a load sensor configured to detect a weight of the cylinder at predetermined time intervals; a temperature sensor configured to detect a temperature local to the cylinder at the predetermined time intervals; and a processing unit configured to: receive weight signals and temperature signals from the load sensor and temperature sensor respectively; and determine, based on the received weight and temperature signals, the status of the gas cylinder; and provide an indication of the status of the gas cylinder to a user. A system for managing deployed cylinders is also provided. Methods and computer readable mediums are also provided.

Claims

1. A system for determining a status of a gas cylinder, the system comprising: a plurality of cylinders; a load sensor configured to detect a weight of the plurality of cylinders at predetermined time intervals; a temperature sensor configured to detect a temperature local to the plurality of cylinders at the predetermined time intervals; and a processing unit configured to: receive weight signals and temperature signals from the load sensor and temperature sensor respectively; determine, based on the received weight and temperature signals, the status of the plurality of gas cylinders; and provide an indication of the status of the plurality of gas cylinders to a user, wherein the load sensor is associated with only one of the plurality of gas cylinders.

2. The system according to claim 1, further comprising a transceiver configured to receive signals from the load sensor and the temperature sensor, the signals corresponding to weight data and temperature data respectively, and send the weight data and temperature data to the processing unit.

3. The system according to claim 1, wherein the processing unit is configured to retrieve a tare weight, calculate a difference between the tare weight and the received weight data to determine a weight difference, the processing unit subsequently configured to: compare the weight difference with a threshold value; and modify the comparison using the received temperature data.

4. The system according to claim 3, wherein the processing unit is configured to estimate a percentage of gas in the cylinder from the modified comparison.

5. The system according to claim 4, wherein the processing unit is further configured to apply a regression model to a change in the estimated percentage of gas in the cylinder over a plurality of time intervals, the output of the regression model indicating the status of the cylinder.

6. The system according to claim 5, wherein the processing unit is further configured to apply a regression model to a change in the weight data over a plurality of time intervals, the output of the regression model being used to determine the status of the cylinder based on the temperature data.

7. (canceled)

8. The system according to claim 4, wherein the processing unit is configured to estimate when the status of the cylinder is considered to be no longer yielding by comparing a percentage of gas in the cylinder with a minimum threshold value and if the percentage is below the minimum threshold value and remains unchanged for a predetermined time period, provide an indication to the user that the cylinder is no longer yielding gas.

9. The system according to claim 4, wherein the processing unit is configured to estimate when the status of the cylinder is considered depleting by identifying a change in percentage of gas over a predetermined time period, comparing the percentage of gas in the cylinder with a reference value and an upper threshold value and if the percentage is less than both the reference value and the upper threshold value, provide an indication to the user that the cylinder is considered to be depleting.

10. The system according to claim 4, wherein the processing unit is configured to estimate the percentage of gas at a first time period and a second time period and calculate a difference, the processing unit subsequently configured to: compare the difference with a lower reference value and if the difference is substantially equal to the lower reference value, provide an indication to the user that the cylinder is considered to be in a period of inactivity; and compare the difference with an upper reference value and if the difference lies between the upper threshold value and the lower threshold value, provide an indication to the user that the cylinder is considered to be leaking.

11. (canceled)

12. The system according to claim 1, wherein the processing unit is configured to retrieve historical data from a cloud-based server and the determination of the status of the cylinder is carried out on the historical data.

13. The system according to claim 1, wherein the processing unit is remote from and in communication with the load sensor and temperature sensor, the load sensor and temperature sensor being local to the gas cylinder.

14.-15. (canceled)

16. The system according to claim 1, wherein the plurality of cylinders is arranged into at least two groups wherein each group comprises at least two cylinders.

17. The system according to claim 16, wherein the gas cylinder with which the load sensor is associated is itself associated with only one of the at least two groups.

18. The system according to claim 1, wherein the processing unit is further configured to retrieve a certified control weight of the gas cylinder and use the certified control weight to calibrate the step of determining the status.

19. The system according to claim 1, wherein the cylinder comprises a machine readable identification tag comprising unique information to identify the cylinder and wherein the processing unit is configured to receive the identification tag applied to the cylinder and associate the unique information with the received weight data and temperature data.

20. The system according to claim 1, wherein the processing unit is further configured to retrieve a plurality of input parameters comprising the weight data and temperature data, apply a weight to each of the input parameters to generate a set of weighted input parameters and sum the weighted input parameters to identify the status of the cylinder.

21. The system according to claim 20, wherein the processing unit comprises a neural network trained on test data, the neural network configured to receive the input parameters, operate on the input parameters, and output a status of the gas cylinder based on the operation performed on the input parameters.

22. The system according to claim 20, wherein the plurality of input parameters further comprises one or more of: estimated percent of gas in the cylinder at a first time; weight of the cylinder at the first time; cylinder tare weight; opening hours of a property where the cylinder is installed; a depletion status of the cylinder indicating whether the cylinder is currently depleting or currently paused; a time since the last known time the cylinder was depleting; aggregated information about typical weight for the cylinder when it stops yielding gas; weather conditions, including temperature data; and signals from at least one additional local sensor selected from the group consisting of: an ultrasound sensor, an external temperature sensor, an infrared temperature sensor, and a flowmeter.

23. The system of claim 22, wherein the depletion status is determined either: by calculating a rate of change of the received weight data over a time period; or using a set of weighted input parameters comprising: a plurality of smoothed and filtered weight data measurements over a time period; cylinder tare weight; a maximum weight of the cylinder; and a typical depletion rate.

24.-32. (canceled)

33. A method of determining a status of a gas cylinder, the method comprising the steps of: detecting a weight of a plurality of gas cylinders, using a load sensor, at predetermined time intervals; detecting a temperature local to the plurality of cylinders, using a temperature sensor, at the predetermined time intervals; receiving, by a processing unit, weight signals and temperature signals from the load sensor and temperature sensor respectively; determining, by the processing unit, the status of the plurality of gas cylinders based on the received weight and temperature signals; and indicating to a user the status of the plurality of gas cylinders, wherein the load sensor is associated with only one of the plurality of cylinders.

34.-63. (canceled)

Description

BRIEF DESCRIPTION OF DRAWINGS

[0110] Embodiments of the present invention will now be described by way of example only with reference to the following drawings in which:

[0111] FIG. 1 is a schematic view of a typical gas cylinder installation;

[0112] FIG. 2 is a graph illustrating the typical lifecycle of a cylinder;

[0113] FIG. 3 is a block diagram of a status determination system;

[0114] FIG. 4 is a schematic view of a gas cylinder installation and a sensor;

[0115] FIG. 5 is a perspective view of a sensor;

[0116] FIG. 6 is a block diagram of a sensor;

[0117] FIG. 7 is an exploded view of a sensor;

[0118] FIG. 8 is an exploded view of an electronic housing in a sensor;

[0119] FIG. 9 is a flow diagram of a typical gas market workflow;

[0120] FIG. 10 is a flow diagram of “pay per cylinder” workflow;

[0121] FIG. 11 is a flow diagram of another “pay per cylinder” workflow; and

[0122] FIG. 12 is a flow diagram of a “pay per volume” workflow.

DETAILED DESCRIPTION

[0123] Many industries use LPG as their primary energy source. Some use LPG from large tanks and others use LPG from individual cylinders. A typical LPG cylinder installation 2 for use in industry is shown in FIG. 1, which shows a set of LPG cylinders 4 connected to a pipeline 6. The cylinders 4 are split into two pairs, each pair connected in parallel. Here, a pair is an example of the previously described group or array of cylinders. The two cylinders 4 on the left side of the installation 2 are separated from the two cylinders 4 on the right side of the installation 2 by a switch 8. The switch 8 determines which pair of cylinders 4 is in fluid communication with the pipeline 6 and so determines from which cylinders 4 gas is being supplied at any given time. The switch 8 can be manually or automatically controlled, for example automatically pressure controlled. When one set of cylinders 4 are empty on one side of the installation 2, the switch 8 is either manually or automatically activated and switches to the cylinders 4 on the other full side. For example, when both cylinders 4 on the left hand side run out, the switch 8 will disconnect these cylinders 4 and connect both cylinders 4 on the right hand side so that a continuous supply of gas is provided.

[0124] It should be noted that throughout the specification the phrase “cylinder installation” refers to an entire LPG gas set up, including at least one gas cylinder (but more commonly an array of cylinders), pipelines, valves, and a switch (separating the array of cylinders when there is more than one cylinder present).

[0125] The terms “group” and “array” are interchangeable are refer to a plurality of cylinders connected together in parallel so that the gas contained in all the cylinders in the group is used at the same time i.e. all the cylinders in a group deplete at the same rate.

[0126] A cylinder installation may therefore comprise multiple arrays, separated by the switch.

[0127] In a typical cylinder installation 2 the side which is currently depleting, i.e. the side of the installation 2 including the cylinder 4 from which gas is being used to supply the pipeline 6, is known as the active side. The other side of the installation 2 including the cylinders 4 which are not yielding gas to the pipeline 6 is known as the passive side.

[0128] It is common practice to order new full cylinders when the cylinders 4 on one side of the installation 2 are empty or for a provider to replace cylinders periodically. This current workflow is illustrated in FIG. 10. Typically, there is a network 70 of customers who are all actively using cylinders for either private or commercial use. Over time, these cylinders start to become empty 72. The consumer can alert the LPG gas supplier that they are out of gas and require a refill. The supplier then delivers new gas cylinders to the consumer so that they are not left without gas 74. Since not all consumers in the network will use gas at the same rate, the cylinders do not all become empty at the same time. This can lead to long delivery routes 76 as the supplier the consumers requiring new cylinders may not necessarily be located in the same area. There is also the chance that large numbers of consumers run out of gas at approximately the same time 78. The supplier is therefore required to carry large numbers of spare cylinders 80 at any given time so that he can add consumers to his delivery route as he becomes informed of the need to replace a cylinder. As can be seen, this current system is inefficient, results in long delivery times, and requires large volumes of cylinders to be transported around even though they may not all be needed.

[0129] A current attempt to solve this problem is shown in FIG. 11. This workflow differs from that shown in FIG. 10 in that the supplier can collect information about which consumers require cylinder refills, such as the number of cylinders and the consumer's location. The supplier can use this information to wait until all the cylinders within a particular area or cluster are empty 82 so that the empty cylinders can be filled at the same time 84. Whilst this does reduce the total number of deliveries made 86, it leaves some customers without gas for unknown periods of time, which is frustrating for the customer.

[0130] A major problem with the workflows shown in FIGS. 10 and 11 is that a provider cannot plan or predict when any given cylinder is going to run out of gas and need replacing. Empty cylinders 4 are typically determined by either the consumer manually checking the status of the switch 8 to see if a switch has been made from the cylinders 4 on one side of the installation 2 to the other or by the consumer experiencing a lack of service and manually activating the switch 8 to switch the cylinders 4. Typically, the cylinders 4 are transported and delivered to the consumer by the LPG supplier. The number of cylinders 4 installed at the consumer site may vary, depending on the requirements of the consumer. Typically, the number of cylinders 4 installed on site is between two and twelve cylinders. Regardless of the number of cylinders 4 on site, the general setup remains the same. That is, the total number of cylinders 4 will be split equally into two groups, the cylinders 4 of each group being connected in parallel. The two groups are connected to the pipeline 6 via a switch 8 and the switch 8 determines from which side of the installation 2 gas is fed into the pipeline 6.

[0131] When the cylinders 4 have run out of gas and need refilling, either the empty cylinders 4 can be swapped out for a set of full cylinders 4 or the cylinders 4 can be directly refilled by a LPG truck. The process then continues again. When the currently depleting cylinders 4 no longer yield gas, the switch 8 reverts to the now full other side. The system is manually switched or automatic depending on the installed change over technology.

[0132] One complete cycle of the typical process therefore includes the following events: [0133] 1. The active side starts depleting for the first time; [0134] 2. The active side becomes empty and the switch is turned to use gas from the passive side; [0135] 3. The active side now becomes the passive side; [0136] 4. The now passive side is refilled while the now active side starts depleting for the first time (refill may occur at any time after the side is empty depending on delivery time etc.); [0137] 5. The active side becomes empty and the switch is turned again so that the passive side starts depleting for the second time; [0138] 6. The recently emptied side is then refilled.

[0139] The above described events in a typical cycle are illustrated in FIG. 2. As can be seen, there is a sharp increase in the amount of gas between the cylinder 4 being empty and being refilled, as expected. Once the cylinder 4 has been refilled, it may be some time before the cylinder 4 starts yielding gas, which is illustrated as a plateau representing a constant, unchanging amount of gas in the cylinder 4. Once the cylinder 4 starts depleting, the amount of gas in the cylinder 4 starts to steadily decrease, indicated by the slope in FIG. 2. Interruptions in usage appear as brief plateaus in the amount of gas present in the cylinder 4.

[0140] A main problem with current systems is being able to accurately estimate the amount of gas remaining in a cylinder 4 so that the switch 8 in the cylinder installation 2 swaps the cylinders 4 at the optimal time, providing the user with an uninterrupted gas supply. Additionally, current systems do not effectively alert the consumer or suppler to abnormal usage, which might indicate the presence of potentially dangerous conditions at the gas cylinder 4. Finally, current systems do not predict when cylinders are going to run out of gas, making planning deliveries (including delivery routes and delivery/collection times) difficult. A consumer is therefore generally required to keep multiple spare cylinders on site so that they can replace each cylinder as it runs out, rather than waiting for the next delivery of cylinders which may not be optimally scheduled.

[0141] The present invention provides a detection system 18 which is able to more accurately determine the amount of gas in a cylinder 4 and its current status, which leads to a more optimal system and better consumer awareness of the current status of the gas cylinder 4.

[0142] A general status determination system 18 for determining normal and abnormal gas usage is shown in FIG. 3. The system 18 includes a gas cylinder installation 2 comprising a plurality of gas cylinders 4 and a sensor 10. The sensor 10 includes a load sensor 20 which is configured to detect the weight of the gas cylinder 4.

[0143] The sensor 10 is in wireless communication with a processing unit 22, which is part of a computer system and remote from the cylinder installation 2. The remote central processing unit 22 includes a processing module 24 which is configured to receive signals from the sensor 10. The signals received from the sensor 10 include weight measurements. The weight of the cylinders 4, which is proportional to the amount of gas present in the cylinders 4, is used to determine when a switch between one side of the installation 2 and the other side of the installation 2 should be made. The weight of the cylinders 4 is therefore monitored, via the sensor 10, during usage. The side of the installation 2, or the cylinder 4, whose weight is being monitored is known as the monitored side. The unmonitored side is the other side of the installation 2, or the other cylinder 4, whose weight is not being monitored. The processing module 24 is configured to determine the volume of gas in the cylinder 4, based on the received weight measurements.

[0144] The processing module 24 is also in wireless communication with a cloud-based server 26, which comprises a memory 28 in the form of cloud storage and a database. The database can include historical consumer data as well as other forms of data, which will be explained in more detail later. The processing module 24 sends the volume of gas in the cylinder 4, along with data relating to the time the determination was made, to the cloud-based server 26 for storage. The processing module 24 is also in wireless communication with a user interface 30. The processing module 24 additionally sends the volume of gas left in the cylinder 4 to the user interface 30. The user interface 30 is configured to indicate to the supplier the volume of gas remaining in the cylinder 4, facilitating automatic ordering of new gas cylinders 4.

[0145] The system is also operable to estimate the status of the unmonitored cylinders from the status of the monitored cylinder over time so as to identify that the unmonitored cylinders also need replacing. This is possible because once the system has determined that the monitored side has gone from a “full” status to an “empty” status, any further reduction in weight, as measured by the load sensor, must correspond to the unmonitored side depleting gas. Since, the unmonitored side will have started from a “full” status, the system is then able to determine when the unmonitored side becomes empty, as the weight will stop decreasing once the cylinders are empty.

[0146] In general, when the monitored side is empty, the system assumes that the unmonitored side will start depleting, and consequently new full cylinders are required on the monitored side. Conversely, when the monitored side is full and starts depleting, the system knows that the unmonitored side must be empty, as the switch has been activated. It is this combination of events which indicates that the cylinders on some side (either the monitored or the unmonitored side) need to be refilled.

[0147] Briefly, in use, the sensor 10 is installed and placed under one of the cylinders 4 within the installation 2, on either of the two sides of the cylinder installation 2, as illustrated in FIG. 4. The sensor 10 monitors the weight of the cylinders 4 on one side of the installation 2 and when it is determined that a first set of cylinders 4 is empty, the switch 8 will be manually or automatically changed from the active side that has just been emptied to the other, passive side which contains full cylinders 4. By monitoring the change in gas levels of the cylinder 4 via the sensor 10, the determination system 18 can inform the supplier that the cylinders 4 on one side of the installation 2 are empty and need to be refilled or exchanged. The LPG supplier can then plan their filling routes based on how much gas their customers have left. Thus, the information from the sensor 10 is used to determine when the gas cylinder 4 will be empty to ensure that a new cylinder 4 can be ordered in time.

[0148] One complete cycle of the process in an installation comprising one sensor therefore includes the following events: [0149] 1. The active monitored side starts depleting for the first time; [0150] 2. The active monitored side becomes empty and the switch is turned to use gas from the passive unmonitored side; [0151] 3. The active side now becomes the passive side, still being monitored; [0152] 4. The now passive monitored side is due to be refilled while the now active unmonitored is being depleted for the first time; [0153] 5. The active unmonitored side becomes empty and the switch is turned again so that the passive monitored side starts depleting for the second time; [0154] 6. The monitored side then starts depleting while the unmonitored side is refilled.

[0155] The status determination system 18 therefore remotely monitors changes in the level of gas in the cylinders 4 and informs the supplier when the cylinders 4 are about to run out, as will be explained in more detail below.

[0156] The sensor 10 of the determination system 18 can be more clearly seen in FIGS. 5 and 7. The sensor 10 includes a load cell 20 and a central processing unit (CPU) or microcontroller 32 enclosed within a casing 34. The casing 34 is made from any material that is hardwearing and easy to manufacture, for example plastic. The casing 34 comprises two halves 34a, 34b joined together using any suitable fastening means 36, for example rivets. The two halves 34a, 34b are sealed against the environment by a sealing member 38.

[0157] The load cell 20 is a weight sensor 20 which weighs the gas cylinder 4. The weight sensor 20 is configured to monitor the weight of the gas cylinder 4 at predetermined time intervals, for example every 10 minutes. At each predetermined time interval, the sensor 20 detects the weight of the cylinder 4 and stores this information locally on a memory 40 within the sensor 10. At the end of a predetermined time period, for example at the end of each day, the signals from the weight sensor 20 that have been collected and stored are wirelessly transmitted to the remote processing unit 22.

[0158] It should be noted that the cylinder installation 2 only includes one sensor 10 that is placed under one cylinder 4 of the installation 2. It does not matter under which cylinder 4 the sensor 10 is placed, or on which side of the installation 2 the sensor 10 is placed. Although the sensor 10 is only placed under a single cylinder 4, it is able to estimate the weight of all the cylinders 4 in the installation 2. This is because the total weight of each cylinder 4 when full is known and so the total weight of all the cylinders 4 in the installation can be determined. In addition, because the cylinders 4 are connected in parallel, rather than in series, the gas usage will be spread equally among the cylinders 4 on the active side of the installation 2. As the sensor 10 knows that only one side is active at any given time, and therefore that only one side of the installation 2 is depleting, the sensor 10 knows that gas usage corresponds to a reduction in weight of the cylinders 4 on the active side of the installation 2. As each side is connected in parallel, all the cylinders 4 on one side will run out at the same time. Thus, once the weight of the cylinders 4 reaches half the initial total weight of the cylinders 4, the sensor 10 knows that the active side has finished depleting and a switch needs to be made.

[0159] In order to send the information from the sensor 10 to the processing module 24, the load cell 20 and memory 40 are connected to a transceiver 42 which sends signals from the load cell 20 and memory 40 to the processing module 24 in the remote processing unit 22. The weight sensor is therefore in data communication with the central processing unit 32.

[0160] The sensor 10 also includes a printed circuit board (PCB) 31 comprising the CPU 32 and an integrated temperature sensor 44, as illustrated in FIG. 6. The PCB 31 and temperature sensor 44 are contained within a housing 46, as illustrated in FIG. 8. The housing 46 comprises a lid so that the electronic components, including the PCB 31 and temperature sensor 44 can be accessed if needed. Other sensors which may also be present, but are not shown, include an independent thermo-sensor or ultrasound sensor. Similarly, the sensor 10 may be in short range or direct communication with local sensors placed around the housing of the cylinders such as local temperature sensors or flow meters.

[0161] The temperature sensor 44 is configured to determine the temperature within the sensor housing and optionally of the atmosphere surrounding the gas cylinder 4 i.e. the ambient temperature. The temperature sensor 44 is therefore configured to detect and determine a temperature local to the gas cylinder 4. The temperature sensor may also be configured to determine the temperature of the cylinder itself by attachment to the cylinder or by close proximity to it. If the sensor is an infra-red temperature sensor as is well known in the art, for example, or a remote thermocouple then it may be possible to accurately determine the temperature of the cylinder itself. The system may also include multiple temperature sensors to input temperature data as parameters. For example, there may be an integrated temperature sensor in the PCB housing, and there may be an additional temperature sensor specifically configured to determine cylinder temperature, cylinder housing temperature, gas temperature, etc.

[0162] Measuring the temperature of the gas cylinder 4 is important because the pressure inside the cylinder 4 is proportional to the temperature. Due to the high pressure inside the cylinder, the LPG is stored as a liquid. When an outlet valve on the cylinder is opened, the pressure is released and the LPG leaves the cylinder as a gas. The higher the pressure inside the cylinder, the more gas is released. The ambient temperature surrounding the cylinder will therefore affect the pressure inside the cylinder. For example, warm surroundings will warm up the cylinder which in turn will heat up the contents of the cylinder, increasing the pressure. As the temperature decreases, so does the pressure inside the cylinder 4 and so at cold temperatures the cylinders 4 will yield less gas. Conversely, in warmer temperatures the pressure inside the cylinder will increase and sob each cylinder 4 will yield substantially all of the gas in the cylinder 4. The amount of available gas in the cylinders 4 is therefore temperature dependent. The status of the gas cylinder is therefore determined using both the detected weight of the cylinder data and a detected temperature local to the cylinder.

[0163] The temperature sensor 44 is configured to monitor the temperature surrounding the cylinder 4 at predetermined time intervals. These time intervals are the same as the time intervals used to monitor the weight of the cylinder 4, for example every 10 minutes. Optionally, the temperature may be measured less frequently as it is less susceptible to change, than the weight of the cylinder, over time. At each time interval, the sensor 44 detects the temperature of the atmosphere surrounding the cylinder 4 and stores this information locally on the memory 40 in the sensor 10. At the end of a predetermined time period, the signals from the temperature sensor 44 that have been collected and stored are transmitted to the remote processing unit 22. The predetermined time period is the same as the time period used to send the weight signals. Thus, the weight and temperature signals are monitored and sent at the same time.

[0164] In order to send information from the temperature sensor 44 to the remote processing unit 22, the temperature sensor 44 is also connected to the transceiver 42 which sends signals from the temperature sensor 44 and the memory 40 to the processing module 24 in the remote processing unit 22. The temperature sensor 44 is therefore in data communication with the remote processing unit 22.

[0165] The transceiver 42 in the sensor 10 wirelessly communicates with the remote processing unit 22. The transceiver 42 sends signals using any readily available network connection, for example using General Packet Radio Service (GPRS) within the Global System for Mobile (GSM) communication system. However, any other communication system that is able to communicate with GSM units can be used.

[0166] The remote processing unit 22 receives signals from the sensor 10, including signals from the weight sensor 20 and temperature sensor 44, and stores the received signals on a memory or database as appropriate. The received signals correspond to measurements taken by the sensors and therefore include weight data and temperature data.

[0167] The processing module 24 retrieves a tare weight of the cylinder 4 from the cloud-based server 26 and, based on the tare weight, the remote processing unit 22 then processes the received data signals, using the processing module 24, to accurately determine the volume of gas present in the cylinder 4. The processing is performed using a number of different algorithms which may for example utilise a neural network, as will be explained in more detail later. The calculation may be based on the LPG weight rather than tare weight. The LPG weight may always the same for a specific cylinder type (e.g. 47 kg). The total weight=tare weight+LPG weight, so we estimate the tare weight by tare weight=total weight−LPG weight.

[0168] It should be noted that the tare weight (including additional weight from cables, valves, etc.) is estimated automatically every time a cylinder is refilled. This has the advantage that the tare weight does not need to be known beforehand, because it can instead be estimated. Since each cylinder is typically of a fixed volume, due to all cylinders being generally manufactured to the same standards and specifications, it is possible to know how much gas is contained within a full cylinder. As explained above, the tare weight can then be estimated by subtracting the fully loaded gas weight from the total weight of the cylinder.

[0169] In some cases it is possible to acquire the tare weight of the cylinder from the distributor. As different cylinders may typically vary in raw tare weight values, even for cylinders which are manufactured to the same specifications, the precise tare weight of the cylinder can be acquired from a distributor, rather than estimating the tare weight, which takes into account any variations between cylinders. The use of an ID tag present on the cylinder can help in this respect because each cylinder can be uniquely identified using its tag meaning that the correct tare weight can be acquired from the distributor.

[0170] In general, the algorithms take as part of their input the weight data from the weight sensor and the time the measurement was taken. As the weight data is monitored over an extended time period, it is possible to calculate how the weight of the cylinder changes with time. That is, the derivative of the weight data can be calculated, using the processing module 24 or central proceeding unit, to estimate a change in weight with time. This can be thought of as a velocity vector. Similarly, the derivative of the change in weight over time can also be calculated, using to processing module 24, to estimate the rate of change of weight over time. This can be thought of as an acceleration vector.

[0171] Depletion can be thought of as a change in the weight of the cylinder over time, and so is comparable to a velocity vector. The rate of depletion is the rate of change of the weight of the cylinder and so is comparable to an acceleration vector. The weight data, velocity vectors, and acceleration vectors can be used as inputs for the algorithms. These definitions will be used throughout the description when describing the status determining algorithms.

[0172] Generally, the algorithms detect how much gas has been used, whether or not the cylinders have been emptied, and whether or not the switch 8 has been activated and changed sides. The remote processing unit 22, which is remote from the cylinder installation 2, therefore monitors the consumer's usage so that it can determine the difference between normal and abnormal usage. Abnormal usage can be used to predict events such as gas leaks, theft, faults in the sensor 10, or user errors, as will be explained in more detail later. When abnormal usage is detected, the processing module 24 sends a wireless signal to the local sensor 10 at the cylinder installation 2 and an alarm 50 in the sensor 10 is triggered. The alarm may be represented by an audio device that could act as an alarm and present a warning of error. Alternatively an alarm may be triggered and sent to a remote warning system.

[0173] The results of the algorithms are then reported to the consumer and the supplier so that they know how much gas has been used in a particular time period, for example during each day, and if replacement cylinders 4 need to be ordered.

[0174] In general the user can track their own usage and benchmark against other users. They can see when they have higher usage than normal. This also makes the end user sure they are not a victim of fraud or wrong cylinder swapping.

[0175] In particular, once the amount of gas remaining in the cylinder 4 has been calculated, the processing module 24 sends the result of the calculation to the user interface 30, which is part of a computing device. This can either be a mobile computing device or a desktop computer in a computer system. The user interface 30 can be a consumer app and/or a supplier/distributor app. That is, both the consumer and the supplier may each have their own user interfaces 30 which can receive the results from the processing module 24. The supplier is therefore automatically informed about how much gas is left in the cylinders 4 in the installation 2 on the local consumer site and can order new cylinders 4 before the consumer runs out of gas. Advantageously, the consumer does not have to manually order the refiling or exchanging of the cylinders 4. In addition, the LPG supplier can better plan their delivery routes and reduce logistics costs. Automatically informing the consumer about how much gas they have left on site allows the consumer to plan their gas usage more effectively and take into account any unexpected peaks in usages, for example during popular business times.

[0176] The problem of determining the optimal time when a refill should be undertaken in a cylinder installation 2 (with one local sensor) can be decomposed into two sub-problems. These sub-problems are detecting when a cylinder 4 has started depleting and detecting when a cylinder is empty. The algorithms used to detect these two main events will now be described. From accurately identifying these two events, the point in the cycle of the set of cylinders can be determined.

[0177] Starting from an installation 2 comprising full cylinders 4 of gas, the first event in the typical gas cycle is depletion. The start of a depletion event occurs when a cylinder 4 starts to yield gas for the consumer to use either the first time the cylinder 4 is used or after a pause in usage. The most important depletion events to be able to detect are the depletion events that occur after the cylinders 4 have been refilled because this indicates that a switch between two cylinders 4 has taken place.

[0178] There are three main techniques which are used to detect depletion events. These techniques are applied sequentially in order to provide a more accurate detection of the depletion event.

[0179] The first technique is a simple check as to whether or not the total amount of gas remaining in the cylinder 4 is significantly lower than the maximum weight of gas that the cylinder 4 can hold. In this context, significant is defined as a difference that is large enough to eliminate the possibility of false positives from noisy data. The first check is therefore a simple comparison to see whether the total amount of gas is lower than a threshold value. If the total amount of gas remaining is significantly lower than the maximum weight of the cylinder then it is assumed that the cylinder has started depleting. At this point there is also a check to determine whether a minimum amount of gas has depleted before any other algorithms are allowed to proceed. This ensures that the algorithms are not performing unnecessary computations or performing computations on incorrect data. This technique is a variant of an inverse clamping mechanism. That is, the event is instantly rejected if the value is below a minimum threshold value and instantly accepted if the value is above a maximum threshold. These thresholds are initially set at a starting value and then determined using machine learning algorithms, so that the accuracy of the algorithms is improved over time.

[0180] In more detail, before the main depletion detection algorithms are run, a number of initial checks are performed. Firstly, it is checked whether the current weight of the gas cylinder 4 is above a maximum threshold value. This maximum threshold value is used to indicate that the cylinder 4 is full and so is chosen to approximately represent the weight of a full cylinder 4. This threshold value can be set to any reasonable value that would indicate the cylinder 4 is full and can be reconfigured and adjusted at any time. As an example, the maximum threshold could be set to 97% of the estimated total weight. If the current weight is above this maximum threshold, it is determined that the cylinder 4 is substantially full and that depletion has not yet begun. The rest of the depletion detection algorithm is then aborted to save computing resources and reduce unnecessary calculations.

[0181] If the current weight of the cylinder 4 is less than this maximum threshold, it is then checked whether the current weight is below an upper threshold. This upper threshold value indicates that current weight of the cylinder 4 is slightly less than the weight of the cylinder 4 when full suggesting depletion has started. This threshold value can be set to any reasonable value and can be reconfigured and adjusted at any time. As an example, this threshold value could be set to 90% of the estimated total weight. If the current weight is below this threshold, it is determined that depletion has started.

[0182] Detecting depletion in the weight range between the upper minimum threshold (for example 90%) and the maximum threshold (for example 97%) can be done using regression, for example linear regression, which attempts to fit a linear curve to the weight data falling within this range. The fit represents the rate of change of the weight data i.e. the acceleration. If the resulting fit has a slope within an interval of acceptance and a R.sup.2 value above a threshold value, for example 95%, it is determined that the data is declining steadily close to the estimated rate of depletion and so it can be assumed that depletion has started. The interval of acceptance and R.sup.2 threshold value are predetermined and can be reconfigured at any time.

[0183] In an example, the expected value used is 0.03, i.e. 30 grams per minute. The interval of acceptance is 0.03±0.015. The system may also work using machine learning, so that individual restaurants expected depletion rates can be used in place of the static value of 0.03.

[0184] If the linear fit does not produce definite results which clearly indicate depletion, the algorithm then tries to find an acceleration outlier within the data. If one or more outliers are found, the algorithm selects the first outlier and then splits the data into two sets. One of the data sets contains this first outlier and the other data set does not contain any outliers. The latter set, i.e. the data set without the outlier, is then selected and the same procedure described above, using regression, is then applied to this data set.

[0185] A second example technique used to detect depletion has begun also uses regression, for example linear regression. The data is divided into slices, each slice representing a fixed, predetermined time period. A linear curve is fit to the most recent data slice. The time period used for the data slice is any reasonable time period but should not be so large such that changes in depletion rate would not be able to be detected and go unnoticed. If the linear curve can be fitted with an acceptable R.sup.2 value, which is a measure of how close the data points are to the fitted regression line, and additionally the slope of the curve is close to the typical depletion rate of the system, this indicates that the cylinder 4 in the system has started depleting. The acceptable value may be set a useful value such as 95%. Further the parameter may be determined by doing a parameter search akin to Hyperparameter optimization.

[0186] A third example technique looks at larger points of negative accelerations, with associated negative velocities, in the curve. If a larger negative acceleration is found, and the slope of the fitted curve after the point where the negative acceleration occurs is also negative, this further indicates that the system has started depleting. The data points after the negative acceleration point is then fitted against a linear curve as before and accepted under the same conditions as before.

[0187] A number of different parameters are used to determine whether a depletion event has occurred. These parameters include the recent weight of the cylinder 4 (typically, the data points will span a 6-12 hour time period), the tare weight and maximum weight of the cylinder 4, the value of a typical depletion rate, and the minimum and maximum values used in the inverse clamping mechanism mentioned previously.

[0188] Machine learning algorithms, such as neural networks, can be used to determine the start of a depletion event. An example of a machine learning algorithm that can be used to detect whether depletion has started is a regular, dense neural network that has been trained on real test data. The neural network ingests smoothed and filtered data and the parameters of the cylinder and commercial property, if relevant, and then uses these to determine whether or not a depletion event has occurred. In particular, the neural network takes as its input vectors of derivatives, i.e. velocities, and accelerations. This algorithm is able to detect when one side of the cylinder installation 2 is empty. When a cylinder 4 starts depleting, it follows that the other side has finished depleting and is now empty and a switch has just recently been made.

[0189] In the present description, we refer to neural networks as an example machine learning algorithm to implement the determination techniques. Where this term is used, it can be considered that any suitable machine learning technique may be used. For example, artificial neural networks may be preferable but similarly a support vector machine, recurrent neural network, or convolutional neural network may be used including for example Tensorflow libraries. Additionally, reinforcement algorithms may be utilised to improve determination.

[0190] In summary, a neural network is trained using parameters such as the ambient temperature of the cylinder at the time test weight data was recorded with the exact volume of gas in the cylinder. The test data may for example use certified control weights approved by regulatory bodies. Once trained and the hidden layer has its weights set, in use the input layer may include the parameters such as the tare weight and acceleration of weight data. Each parameter is therefore weighted by the trained neural network and summed to derive a conclusion as to the likely event. The output layer is the decision of the event categorisation as a result of the weighted parameters. Advantageously, the neural network can be trained on data from more than one user, improving the accuracy of the neural network over time.

[0191] In a typical lifetime cycle of one cylinder 4, the depletion of gas will likely start and stop several times as the consumer does not generally need a constant supply of gas. For example, the consumer will use the gas during commercial business hours and will not use the gas during the night. There are therefore several start and stop events which take place during the complete cycle of the cylinder 4.

[0192] In order to detect a temporary pause in depletion the same algorithm that is used to detect the main initial depletion event is used, except that the initial conditions are not used. In addition, the algorithm reverses the requirements for the slope, requiring in this case steady non-declining data, instead of steady declining data.

[0193] In order to determine when the cylinder has temporarily started depleting again after a pause, the same algorithm that detects when the initial, beginning depletion event occurs is used however, in this case, the initial conditions are not used. This is because the cylinder is resuming depletion after a pause has taken place and so the initial conditions will vary each time the cylinder restarts depleting.

[0194] It is useful to be able to predict when the cylinder 4 is about to become empty, or when it will no longer yield gas, before this state is actually reached so that replacement cylinders 4 can be ordered before the consumer actually needs them. An emptiness predication algorithm is therefore used to predict when this point will occur. In this context, “empty” refers to the ability of the cylinder 4 to yield gas and not the amount of fluid in the cylinder 4, in either liquid or gas form.

[0195] Thus an empty cylinder 4 is a cylinder 4 that no longer has the ability to yield gas and an emptiness prediction algorithm predict when this state will be reached. Factors that affect the amount of gas a cylinder can yield include gas volume, temperature, and pressure due to a state change of the LPG in the cylinder.

[0196] The emptiness prediction algorithm retrieves the current rate of depletion up until a particular point in time and then extrapolates the rate of depletion to provide an indication of when the cylinder 4 will become empty. Depletion rate estimates are calculated at various different time intervals (for example over the last 24 hours, over the last week, and over the life time of the cylinder) which can be combined to provide a more accurate result. An accurate prediction of the end point will allow the gas suppliers and distributors to plan deliveries in advance on a much larger scale.

[0197] The emptiness prediction algorithm is also a machine learning algorithm which uses a regular, dense neural network to perform the extrapolation and predictions. The inputs to the neural network include parameters such as “x % gas remaining” or “n days since depletion started”. The output of the algorithm is a number estimating how much longer the cylinder 4 can be used for, based on current usage rates. This result may be presented, for example, as a number of remaining days or hours.

[0198] Data for specific locations or specific times of the year can also be input into the emptiness prediction algorithm. For example, the neural network may also use parameters such as “the average depletion rate on a Monday” and “average depletion rate during Easter” for a particular restaurant. These factors are combined with parameters common to all cylinders 4, such as “percentage of gas remaining” to estimate when the amount of usable gas in the cylinder 4 will be reached. Thus, an advantage of using a neural network over current systems is that the neural network learns and is trained on data from all users that are part of the system, improving the accuracy and performance of the neural network over time.

[0199] The final event in the lifetime of a cylinder 4 is when the cylinder 4 becomes empty and the switch 8 in the cylinder installation 2 switches cylinders 4 on the other side of the installation 2. In order to determine when a switch between cylinders 4 should be made, it first needs to be determined when a cylinder 4 has been emptied. As mentioned, this can be difficult because a cylinder 4 can be physically non-empty but still not yield any gas to the pipeline 6. Thus, instead of detecting an absolute value of “emptiness”, the status determination system instead detects when the cylinder 4 has ceased to release any more gas i.e. when the cylinder 4 has reached a steady, non-depleting state.

[0200] Furthermore, although a cylinder 4 may have stopped yielding gas, this does not necessarily correspond to an “empty” condition. It could instead indicate a temporary pause in gas usage at the consumer end and so the algorithms need to be able to identify and distinguish between these two situations.

[0201] Various factors can affect whether or not a cylinder 4 will completely empty during one complete cycle or if instead the cylinder 4 will still retain some residual fluid. One factor of particular importance is the temperature which is proportional to the pressure inside the cylinder 4. In cold temperatures the cylinders 4 will have some residual fluid remaining inside at the end of their cycle and in warmer temperatures, each cylinder 4 will be substantially empty at the end of each cycle.

[0202] Other external factors that affect the amount of usable gas in each cylinder 4 include the air pressure. Thus, in order to accurately predict the emptiness of cylinders 4, local weather data can be combined with the weight and temperature data to provide a more accurate estimation of the amount of gas present in the cylinder 4. The weather data can be stored on the cloud server 26 and retrieved by the remote processing unit 22, via wireless data communication systems. The retrieved air pressure is therefore an additional weighted parameter, in conjunction with the temperature and weight parameters, used by the status determining algorithms

[0203] As can be seen, a number of parameters are used to determine when the cylinder 4 has reached an empty state rather than a temporary pause. As well as the above mentioned parameters, additional parameters include the current percentage of gas in the cylinder 4, current weight, and current cylinder tare weight. The status of the cylinder 4 at a time interval immediately before any current measurements are taken by the sensor 10 can be used to predict the current status of the cylinder 4. For example, the system 22 can take into account whether the cylinder 4 has been previously depleting, in which case it may now be empty, or whether it was temporarily paused, in which case it is likely still paused.

[0204] If the cylinder 4 is located at, and used on, commercial property, the opening hours of the property can be taken into account to help determine whether the cylinder 4 is empty or whether there is a temporary pause in usage.

[0205] Other information which may also be taken into amount are the time since the last known time the cylinder 4 was depleting and aggregated information about the typical remaining weight of the cylinder 4 when it stops yielding.

[0206] The algorithm used to detect that a cylinder 4 is empty works by directly inspecting the data and setting a threshold at which the cylinder 4 is considered empty, for example 5% of the total weight of the cylinder 4. The threshold can be set at any reasonable value, and is reconfigurable by the supplier. In conjunction with this algorithm, the data is also inspected to make sure that depletion has stopped rather than is temporarily paused.

[0207] When the emptiness algorithm is combined with contextual information, as described above, the emptiness algorithm becomes a machine learning algorithm. The primary contextual information is the ambient temperature which is measured using the temperature sensor 44. As explained, if the temperature of the cylinder 4 is sufficiently low the cylinder 4 can stop yielding gas before it is physically empty. This is because a lower temperature corresponds to a lower pressure and so the gas cannot escape the gas cylinder. Correlating this information with local weather reports provides similar contextual information from another source. In an example, the local weather report is another source of contextual information regarding the temperature. This could make inferences using temperature more accurate (two is better than one, and so on). Weather data may also relate to the prediction of potential temperature drop and drop of pressure. The machine learning emptiness algorithm is a regular, dense, neural network trained on real test data. The network takes as its input a vector of the percentage values of the amount of gas in the cylinder 4 along with the contextual information from, for example, the temperature sensor 44.

[0208] Once it has been determined that the cylinders 4 on one side of the installation 2 are empty, a switch is made and the other cylinders 4 start yielding gas. The whole process then begins again, the sensors 20, 44 monitoring the weight and temperature of the cylinders 4 and the processing module 24 calculating the amount of gas in the cylinder 4 using the machines learning algorithms.

[0209] At the same time the switch is made, the supplier is notified of the switch and the empty cylinders 4 can be refilled. It is important to be able to detect once the cylinders 4 have been refilled so that the switch 8 does not switch to an empty set of cylinders 4. In this case, the consumer or supplier can be alerted, via the user interface 30, that there is a problem with the installation 2.

[0210] In order to detect that a cylinder 4 has been refilled, after initially filtering the weight data, an algorithm is run on the weight data which attempts to find a derivative which is both positive and of at least a target minimum value. The minimum value can be set to any reasonable value, for example 10 kg. The algorithm can be a regular, dense, neural network that has been trained on real test data. The neural network takes as its input a single vector of derivatives and outputs a Boolean flag. The determination system 18 automatically deduces the new total weight of the gas cylinder 4 which is being monitored. Given that the total amount of initial gas in each cylinder 4 is fixed, and therefore known, it is possible to deduce the tare weight of the cylinder 4. This information is then used to accurately calculate the remaining percentages of gas in the cylinder 4 as the cylinder 4 is being used. When a cylinder 4 has been refilled the system 18 will mark the customer as no longer needing new gas. This detection algorithm therefore forms part of a subsystem which informs the gas distributors which customers need gas.

[0211] As well as monitoring normal usage and operation of the gas cylinders 4, it is important to be able to determine when the gas cylinders 4 are not behaving as expected or when consumer usage appears abnormal. Unusual behaviour of the gas cylinders 4 may indicate that the gas cylinder 4 is leaking which could be dangerous. Abnormal usage by the consumer could indicate that the cylinder 4 has been stolen or is being used by unauthorised persons. Other, non-standard events which are important to be able to identify include user errors, installation errors, and sensor failures. These events can all be indicated to the consumer and supplier through the user interface 30 and the local alarm 50 at the cylinder installation 2.

[0212] Algorithms for detecting abnormal usage and abnormal operation of the gas cylinders will now be described.

[0213] Detecting whether or not a leak has occurred is important so that potentially hazardous situations can be avoided. In order to detect whether or not the cylinder 4 is leaking a leak detection algorithm first detects when a cylinder 4 has stopped depleting “for the night”. That is to say at the end of a predetermined time period, for example at the end of each day, the consumer will no longer be using the gas cylinder 4 and so there will be a temporary pause in depletion. The leak detection algorithm then monitors the status of the cylinder 4 and determines when depletion has begun again, indicating the start of the next time period, for example the start of the next day. A linear curve is then fit to the sensor data between the depletion stopping and starting up again using regression, typically linear regression, and the average rate of depletion between the two time periods, for example over the night, can be calculated by analysing the slope of the data. If the slope is within a given threshold and has a very high R.sup.2 value, the data is flagged as leaking. In particular, if the slope is above a lower bound which indicates a period of inactivity, for example a lower bound substantially close to zero, but the slope is also below an upper bound which indicates usage, any slope falling within this range will indicate that there is a leak. This is because a substantially non-zero slope indicates depletion whilst a substantially horizontal slope indicates there is no depletion. A leak occurs when small levels of gas escape from the cylinder 4 but not at the same levels as would be used by a consumer 4. The upper and lower bounds are therefore chosen to detect this small change in gas levels.

[0214] Thus in general in order to detect a leak, periods of inactivity are analysed to see whether or not there is some small level of depletion happening during the period of inactivity. In this sense, the determination system can be thought of as self-aware because it is able to itself detect if something is not right, this refers to the sensor's capability to itself detect suspicious behaviour of the system. It is also possible for gas distributors to keep a closer eye on the behaviour of the gas cylinders 4 and their usage, allowing them to react more precisely to customer demands.

[0215] The performance of the determination system 18 relies on the sensor 10 at the cylinder installation 2 measuring the weight and temperature of the cylinders 4 and sending these signals to the remote processing unit 22. If the sensor 10 fails then it will stop sending real time data to the remote processing unit 22 and the current status of the cylinders 4 cannot be determined. This is problematic as the supplier relies on accurate estimations of the amount of gas left in the cylinders 4 so that they can predict when the consumer will run out of gas. Abnormal usage will also not be able to be detected if the sensor 10 is not working properly.

[0216] If the sensor 10 fails then historical data, which has been collected over time and stored on the cloud-based server 26 while the sensor 10 was working, is instead used to predict when it is time to change or refill the cylinders 4. Using historical data reduces the impact of a non-functioning sensor 10 and ensures that the consumer will not run out of gas while the sensor 10 is down.

[0217] The historical data may refer to some or all of the data emitted by the sensor available. Some of it might be raw data, some of it might be aggregations. This historical data tells the system about the expected amount of time between refills, between a refill and until depletion, the expected depletion rate, and any other types of feature/event detected.

[0218] When a sensor fails, the system can—using the latest known data of a sensor—thus predict the expected amount of time until a refill has occurred. When a refill then has occurred (which on failure will be set at the expected time, or edited manually by the users of the system), the system may use the historical median difference (the exact technique might vary in the future) between refills as the next expected refill.

[0219] Sensor failure can be detected using various techniques. For example, if the remote processing unit 22 stops receiving sensor data over a predetermined time interval, it will determine that the sensor has failed. In addition, or as an alternative, the sensor may send an error signal to the remote processing unit 22 which, in turn, may send an indication of the sensor failure to the user interface 30 to notify the consumer and/or distributor.

[0220] The system may use two primary techniques to detect a sensor failure. First, if the sensor stops sending data (each sensor has a defined transmission interval, typically around 30 minutes. If a sensor doesn't send data within 2× this interval, the sensors state is defined as _not sending data). Second, the sensor itself sends some error signal. The sensor can send a number of different signals indicating the specific error that might have occurred.

[0221] When the sensor 10 fails an alarm is triggered. This alarm alerts the gas supplier that the sensor 10 is not functioning properly and needs to be replaced the next time the gas supplier is at the customer site. The remote processing unit 22 will then use historical data, which includes a safety margin, to predict when it is time to do the next gas refill. The supplier then replaces the sensor 10 when doing the next refill and the system 18 resumes as normal for that particular consumer. The safety margin ensures that the cylinder 4 is refilled or replaced slightly too soon rather than too late so that the consumer does not have to go through a period in which they do not have a supply of gas.

[0222] Other errors which are important to identify include user errors, user interference, and data pollution due to environmental reasons. Examples include: Earthquakes (vibrations); Trains and trams (vibrations); General traffic (vibrations); Very high temperatures; Very low temperatures; and, Other objects periodically or otherwise touching the sensor. An error detection algorithm looks at the number of noisy data points that would be removed during a noise removal process. If the number of noisy points is consistently high, compared to a predetermined acceptable number of noisy data points, then this indicates that either the sensor is having problems or there is an external factor which is polluting the data.

[0223] The different types of disturbances can be categorised using machine learning and classifying training data used by the neural network. The machine learning error detection algorithm uses standard case-based reasoning with Bayesian interference to distinguish between user errors or interference, sensor failures, and environmental factors affecting the data.

[0224] In order that the status detection and depleting algorithms perform optimally, it is important to be able to remove and smooth noise. In general, this process involves removing outliers and then smoothing the remaining data points.

[0225] It is important to distinguish between noise removal and noise filtering. Algorithms that perform noise removal are designed to remove points which are considered too noisy to be used in further calculations. These data points are typically one-off outliers or similar. Noise removal algorithms do not alter any point that is not considered an outlier and the overall shape of the data remains the same. On the other hand, noise filtering algorithms smooth the data. These algorithms filter, or smooth, unwanted components or features of a signal. As a result, the overall shape of the data will likely be different to the original shape after noise filtering algorithms have been applied. In general, the removal algorithms are applied first and then the data is filtered.

[0226] Different events require different smoothing and filtering techniques. For example Savitzky-Golay or SOS filtering greatly affects discontinuities in the graph and so these filtering techniques are not suitable for detecting when a refill event has occurred, as this will be indicated by discontinuity in the weight data graph. Since it is the discontinuity feature which is of interest when detecting this particular event, filtering techniques which heavily affect discontinuities cannot be used.

[0227] A typical workflow for a current gas supplier delivering gas cylinders to a consumer is illustrated in FIG. 9. Typically, the customer manually places an order for a gas cylinder 4 through a central call centre 52. This order then gets manually registered in an order management system 54 and the customer's location details will be entered into a route planning system 56. Based on a list 58 of customers that have placed orders, the supplier will plan a route accordingly and deliver 60 the gas cylinders 4. As each cylinder 4 is delivered to a customer the supplier will make a manual record 62 of a successful delivery which can then be entered into the main management system 54 so that there is a log of that particular delivery session.

[0228] As is evident, current systems need the driver and administration to manually register orders into the order management system 54 and record when the deliveries have been completed.

[0229] The proposed determination system 18 described here reduces the need for manually keeping track of when cylinders 4 need replacing, verifying that the delivery has been made, and updating consumer records because the determination system 18 can accurately estimate consumer usage and predict when the gas cylinder will run out. Furthermore, the determination system 18 can flag up abnormal occurrences and alert both the consumer and supplier to potentially hazardous situations. Thus the chance of human error can be reduced.

[0230] A side effect of being able to accurately determine the amount of gas in the cylinder installation 2 and predict when the cylinders 4 will become empty is that the logistics system used by the supplier can be optimized. The logistics system optimization will make use of the prediction data, local sensor data, global condition data (through the use of cloud based sensors and/or remote sensors), and big data stored in the cloud-based server 26.

[0231] An optimized logistics workflow is illustrated in FIG. 12. As with current workflows, there is a network 70 of customers who are all actively using gas cylinders, for either private or commercial use. Again, over time these cylinders start to become empty 72.

[0232] However, unlike with current workflows, the new workflow uses the prediction system 18 to monitor how much gas the consumers have left in their cylinders, as well as being able to alert a consumer when their cylinders are nearly empty rather than discovering they are empty when it is too late. The supplier can therefore plan a delivery route in advance of consumer actually running out of gas 90 so that a consumer does not have to go through a period with no gas. Since the supplier has information about how much gas is left in all the cylinders in the network, the supplier can be geographically selective 92 with his refills by only refilling cylinders within a particular region. This may include refilling consumers that are nearly empty as well as topping up consumers which would become empty in the near future. The supplier therefore needs to make fewer delivery trips 94, as well as being able to carry only the amount of gas required for a particular area. Another advantage of using the prediction system 18 to optimize the delivery workflow is that a supplier can chose to redistribute 86 cylinders between consumers, depending on the level of gas usage for each consumer, to make sure that all the cylinders in one particular area are emptied at approximately the same time. This further improves the efficiency of route planning between consumers and reduces the number of spare cylinders that need to be carried by the supplier.

[0233] The logistics system can therefore be optimized in terms of routing planning (in that similar locations can be grouped together and the same location is not visited more than once per trip), collection and delivery times to better suit the consumer, as well as reduced delivery counts.

[0234] Additionally, with a more optimized system, the number of cylinders that a consumer needs to store on site can be reduced, in some cases halved, as a result of better route planning, and delivery times more efficiently coinciding with times when the cylinders are about to become empty. This may also allow the number of deliveries (i.e. the number of trips that need to be made by the supplier to delivery gas to the consumers) to be reduced.

[0235] Additionally, the determination system 18 applied to a current cylinder logistics system would provide the possibility of selling gas to the consumer based on the amount of gas actually used by the consumer, rather than per unit cylinder. In this case, the consumer only pays for the gas they actually use rather than paying for whole cylinders, which may be replaced or refilled before they are completely empty. As explained above, this is made possible through the use of certified control weights.

[0236] As well as the cost saving advantages to the consumer, this system would reduce the number of cylinders 4 that need to be stored on site. If fewer cylinders 4 are stored at different consumer site locations, the security of the overall LPG provider system is improved because it is easier to store and keep track of fewer cylinders 4. In addition, the impact of an error occurring is also reduced if the total number of cylinders 4 in the system in reduced. Furthermore, the size of each cylinder 4 can be reduced because the determination system 18 is able to more accurately determine how much gas in present in each cylinder 4 and so a smaller safety margin, or safety buffer, of gas in the cylinder 4 is needed before new cylinders 4 are ordered.

[0237] As before, a local weight sensor 20 sends signals to a remote processing module 24 which uses the signals to estimate the amount of gas present in the cylinders 4. Usage patterns that are linked to specific dates along with weather data are used to estimate when consumer usage is likely to increase beyond normal usage. For example, during cold weather more gas will be used in central heating systems and during holidays certain commercial properties such as restaurants will be busier and so will use more gas for catering.

[0238] It would therefore be useful to be able to determine whether or not the gas cylinders 4 received by the consumers are certified and regulated rather than being supplied by fraudulent distributors. Additionally, since gas cylinders 4 are a valuable commodity for the consumers it would be useful to be able to track and monitor the gas cylinders 4 that have been delivered to each consumer to prevent them from becoming targets for thieves. Furthermore, if the cylinders 4 do get stolen, it would be beneficial to be able to track the cylinder 4 so that it can be found and returned to either the consumer or the gas distributor.

[0239] The determination system 18 can therefore include a tracking system for tracking and identifying each gas cylinder 4 that is part of the system. In order to be able to track and identify each cylinder 4 in the overall system, each cylinder 4 may be provided with an ID tag. This could be an RFID tag or a QR code. As well as being able to track the cylinder 4, the ID tag secures the delivery of the cylinder 4 when it is scanned as both the supplier and the consumer will know whether the delivery of the cylinder 4 has been successful or if something may have happened to the cylinder, such as theft, on route. As mentioned previously, the ID tag can also be used to uniquely associated a certified control weight with a particular cylinder. This allows a more accurate payment method to be implemented which is based on the value of gas used by each consumer rather than the number of cylinder's delivered to each consumer, as has been explained previously.

[0240] The sensor 10 may include a tracking system reader 64. This may be, for example, an RFID reader. Advantageously, an RFID reader can reduce the likelihood of human errors which may occur when data from large numbers of cylinders needs to be read and recorded. The tracking system reader 64 is used to read the ID tags on the cylinders 4 and send signals wirelessly to the remote processing unit 22. The signals received at the remote processing unit 22 can then be sent to the user interface 30 which notifies the supplier exactly which cylinders 4 have been distributed to and used by which consumers. The supplier and consumer user interfaces 30 can be in wireless data communication with each other so that information about the cylinder 4, (such as estimated delivery time, cost, current usage, new orders) can be shared between the supplier/distributer and the consumer.

[0241] To verify the safety and security of the determination system 18, the sensor 10 and algorithms in the system 18 can be calibrated using a certified control weight. The certified control weight can be present at a central site or in a transport vehicle. Thus, an accurate certified control weight is used to verify the amount of gas used in the cylinder 4. The certified control weight can also be used to detect errors in the system 18, and to secure and open up payment to the consumer based on gas usage rather than based on number of cylinders 4 delivered. For example, if the certified control weight indicated that a particular cylinder had been under-filled, compared to an expected “full” cylinder weight, the consumer would receive a cylinder with less gas in and so should not be charged the same price as the price of a “full” cylinder. Similarly, if the certified control weight indicated that a particular cylinder had been overfilled compared to an expected “full” cylinder, the consumer can be charged for this extra gas that they have received, rather than the gas supplier losing out on money.

[0242] In particular, by accurately determining and verifying the volume of gas in each cylinder 4 before and after the gas cylinder 4 has been refilled, it is possible for the gas supplier to charge the consumer for the volume or gas that was actually used, rather than charging a fixed price per gas cylinder. This can be implemented in two ways.

[0243] One possibility is to give the consumer a discount on each cylinder 4 which is proportional to the amount of gas that would typically be left in the cylinder 4 when the cylinders 4 are due to be refilled or exchanged. For example, the consumer may receive a 20% discount on each cylinder 4 and when the amount of gas in each cylinder 4 is below 20%, the gas distributor replaces the cylinder 4 with a new full cylinder 4. The customer therefore only pays for the gas that they actually use.

[0244] An alternative solution is for the gas distributor to replace all the cylinders 4 in a geographical region which contain an amount of gas below a certain threshold.

[0245] For example, the gas distributor may replace all cylinders 4 in a particular region which contain less than 40% gas. When the cylinders 4 are returned to a filling station at the supplier end, the economic compensation from the consumer is calculated based on how much gas is left in the cylinder 4. This calculation is done using the certified control weight. Each cylinder 4 can be tracked using, for example, RFID tags so that each cylinder 4 is associated with a particular customer.

[0246] The control weight measures the weight of each cylinder 4 at the cylinder installation 2 when the cylinder 4 is refilled, or at a separate location if the cylinder 4 is refilled at a separate filling station, and this information is then sent to the remote processing unit 22. At the same time, the tracking ID tag on the cylinder 4 is scanned by the tracking system reader 64 and the scanned information is also sent to the remote processing unit 22. The combination of using a certified control weight and a tracking system ensures that each cylinder 4 can be correctly identified and accurately weighed so that the customer can be charged the correct amount based on their actual gas usage.

[0247] Charging customers per volume of gas used rather than per cylinder will reduce the cost of logistics for the gas distributor. This is because the distributor can change their deliveries from being order orientated to being geographically orientated. The gas distributor can now focus on all the cylinders 4 in a particular geographical region. For gas installations, the supplier can also change all the cylinders 4 in a particular region in one delivery. This dramatically reduces the number of deliveries needed at a specific location.

[0248] As is clear, the main users of the status determination system 18 are the gas distributors and suppliers who use the system 18 to monitor how much gas their customers have left before the cylinders 4 need replacing. The whole process can be automated which reduces the risk of human error in the system.

[0249] Accurate usage estimation by the processing module 24, based on signals from the sensor 10, provides a number of advantages. Firstly, the number of user errors is dramatically decreased. Previously, false alarms could be created because of the end-user intervening with the sensor 10 during installation. For example the user may not place the cylinder 4 directly on top of the cylinder which may lead to false weight measurements of the cylinder 4. The presently described system 18, on the other hand, avoids this problem through automation which provides consumers and suppliers with an accurate estimation of the amount of gas left in each cylinder as a percentage of the total amount of gas. As a result, errors can be detected either manually by the gas distributor or automatically through the use of algorithms. Being able to detect user errors avoids the possibility of unnecessary deliveries and so keeps logistics costs down.

[0250] Secondly, abnormal patterns in gas consumption can more easily be detected which may help identify and reduce gas leakages and other potentially hazardous situations.

[0251] False alarms due to low temperatures are also avoided through accurate calculation of the amount of gas left in cylinder.

[0252] In summary, a system is provided which allows users (including gas suppliers and consumers) to monitor and infer the status of an entire cylinder installation by monitoring only one cylinder within the entire installation. There is no longer the need to monitor an entire array of cylinders within the installation, which has cost savings as well as efficiency improvements. The system provided is able to estimate patterns for LPG usage based on “big data”. This can give an optimized logistics system, by knowing the usage pattern of a consumer. For example, data shows low usage on Saturday and higher usage on Fridays etc. which can allow the system to accurately predict usage and future available gas yield.

[0253] Additionally, a more accurate predication system of when the consumer will run out of gas is provided which has benefits for the supplier, distributor, and consumer of the prediction system.

[0254] Finally, the system may also provide energy saving advantages. Since the status and typical depletion rate of the cylinders being monitored is known, the sensors can transmit data less often within certain time periods (straight after refill, for instance), when it is known that the time until the next event is likely to be long. This therefore saves energy of the sensor.