Methods, (measuring) devices, and components thereof, for identifying events in a material-processing or material production process using event patterns

Abstract

A measuring device comprising a sensor for the high frequency detection of a measurement variable, such as sound, structure-borne noise, current, voltage, optical or magnetic measurement values and the like, with an operator network comprising at least one operator. An operator network that includes a notification operator, a machine control operator and/or a difference operator. The devices can be configured to identify events in a material-processing and/or material production process on the basis of a multidimensional data stream obtained during the process and having temporally resolved frequency and energy information, and to the use of defect patterns and/or event patterns and/or energy data, the data being supplied to a pattern operator or multiple pattern operators arranged in parallel or one after the other.

Claims

1. A measuring device comprising a sensor for high-frequency detection of a measurand, such as sound, structure-borne sound, current, voltage, optical or magnetic measurement values and the like, characterized by an operator network according to one of the following claims with at least one operator that includes one or more of a) decision operator for deciding whether a component in a material processing and/or production process or in a processing and/or production device is OK or not OK, and wherein said decision operator having an input for a fault pattern recognition operator, an input for an event pattern recognition operator and/or an input for a free pattern operator, and wherein said decision operator has internal logic and internal parameters for parameterisable and/or programmable linking of the input data, and having an output for an OK or not OK signal, and wherein said decision operator includes a two- or three-dimensional decision field; and/or b) a pattern comparison operator having an input port to receive an input data stream and an output port to output an output data stream and/or value corresponding to a measure of an agreement between a pattern in the input data stream and one of a plurality of patterns from a pattern database, and wherein the pattern comparison operator comprises a two-, three- or multi-dimensional characteristic map with one or more assignment regions.

2. An operator network comprising one, two or more pattern recognition operators, optionally a free pattern operator and at least one decision operator according to claim 1.

3. The operator network according to claim 2, characterized by a visualization operator, a machine control operator and/or a difference operator.

4. The operator network according to claim 2, characterized in that a plurality of operators, in particular pattern operators, are arranged in parallel or cascading form.

5. A mobile device, in particular with independent power supply, in particular a sensor or display module, characterized by being designed as an operator with an input and/or output data stream for embedding in the operator network or in the measuring device according to claim 1.

6. A device for identifying events in a material processing and/or production process on the basis of a multi-dimensional data stream obtained during the process with time-resolved frequency and energy information, using fault and/or event patterns and/or energy data, characterized in that the data pass through one pattern operator or a plurality of pattern operators arranged in parallel or in succession and are supplied to one decision operator or a plurality of decision operators arranged in parallel or in succession according to claim 1.

7. A graphical user interface for creating and/or editing a network of operators, and/or a measuring device comprising a graphical user interface, characterized by graphical positioning and linking of the operators according to claim 1.

8. A method for creating and/or editing the operator network according to claim 1.

9. A decision operator for deciding whether a component in a material processing and/or production process or the processing and/or production device is OK or not OK; said decision operator having an input for a fault pattern recognition operator, an input for an event pattern recognition operator and/or an input for a free pattern operator; said decision operator having internal logic and internal parameters for parameterisable and/or programmable linking of the input data, and having an output for an OK or not OK signal; said decision operator includes a two- or three-dimensional decision field.

10. The decision operator according to claim 9, characterized by three decision fields each having at least one temporally variable boundary line.

11. A measuring device comprising a sensor for high-frequency detection of a measurand, such as sound, structure-borne sound, current, voltage, optical or magnetic measurement values and the like, characterized by an operator network, said operator network includes a pattern comparison operator having an input port to receive an input data stream and an output port to output an output data stream and/or value corresponding to a measure of an agreement between a pattern in the input data stream and one of a plurality of patterns from a pattern database; the pattern comparison operator comprises a two-, three- or multi-dimensional characteristic map with one or more assignment regions.

12. The measuring device according to claim 11, characterized in that the pattern comparison operator identifies event patterns and/or fault patterns.

13. A measuring device for identifying events in a material processing or production process; said measuring device comprising a sensor for high-frequency detection of a measurand and an operator network; said measurand includes one or more properties selected from the group consisting of sound, structure-borne sound, current, voltage, optical measurement values and magnetic measurement values; said operator network includes one or more of I) a decision operator configured to determine whether a component in a material processing and/or production process or in a processing and/or production device is OK or not OK, and wherein said decision operator has an input for 1) a fault pattern recognition operator, i1) an input for an event pattern recognition operator and/or ii1) an input for a free pattern operator, and wherein said decision operator has a) internal logic and internal parameters for parameterisable and/or programmable linking of the input data, and/or b) an output for an OK or not OK signal, and wherein said decision operator includes a multi-dimensional decision field; and/or II) a pattern comparison operator that has an input port to receive an input data stream and an output port to output an output data stream and/or value corresponding to a measure of an agreement between a pattern in said input data stream and one of a plurality of patterns from a pattern database, and wherein said pattern comparison operator includes a multi-dimensional characteristic map with one or more assignment regions.

14. The measuring device as defined on claim 13, wherein said decision operator includes three decision fields that each have at least one temporally variable boundary line.

15. The measuring device as defined in claim 13, wherein said pattern comparison operator identifies event patterns and/or fault patterns.

16. An operator network comprising at least one pattern recognition operator, a free pattern operator, and/or at least one decision operator as defined in claim 13.

17. The operator network as defined in claim 16, further including a visualization operator, a machine control operator, and/or a difference operator.

18. The operator network as defined in claim 16, wherein a plurality of operators are arranged in parallel or cascading form.

19. A mobile device that includes an independent power supply and one or more of a sensor and display module; said mobile device is configured to function as an operator with an input and/or output data stream for embedding in said operator network and/or in said measuring device as defined in claim 13.

20. A device for identifying events in a material processing and/or production process on the basis of a multi-dimensional data stream obtained during said process with time-resolved frequency and energy information by using fault and/or event patterns and/or energy data; said device is configured to enable said data to pass through one pattern operator or a plurality of pattern operators that are arranged in parallel or in succession and are supplied to one decision operator or a plurality of decision operators arranged in parallel or in succession as defined in claim 13.

21. A graphical user interface for creating and/or editing a network of operators, and/or a measuring device comprising a graphical user interface, wherein there is provided graphical positioning and linking of one or more of said operators as defined in claim 13.

22. A method for creating and/or editing said operator network as defined in claim 13.

Description

DESCRIPTION OF DRAWINGS

(1) FIG. 1 illustrates a prior art 1 the data stream recorded as a function of time t, frequency f and intensity I used to evaluate a cutting process.

(2) FIG. 2 illustrates an operator network in accordance with the present disclosure.

(3) FIG. 3 illustrates a fault pattern operator in accordance with the present disclosure.

(4) FIG. 4 illustrates a free pattern operator that can have, in addition to a data stream operator, a time operator and a position determination operator as shown in FIG. 3.

(5) FIG. 5 illustrates a first multi-dimensional decision field in accordance with the present disclosure.

(6) FIG. 6 illustrates a multi-dimensional NOK decision field in accordance with the present disclosure.

(7) FIG. 7 illustrates a multi-dimensional OK decision field in accordance with the present disclosure.

(8) FIG. 8 illustrates a multi-dimensional NO decision field in accordance with the present disclosure.

(9) FIG. 9 illustrates a decision operator which monitors the maintenance condition of the bending device.

(10) FIG. 10 illustrates a non-limiting configuration for a difference operator in accordance with the present disclosure.

(11) FIG. 11 illustrates a non-limiting example of analysis sequences that are created in the graphical operator editor in accordance with the present disclosure.

(12) FIG. 12 illustrates a non-limiting example of a network for pattern analysis and decision making in accordance with the present disclosure.

(13) FIG. 13 illustrates a network that combines data stream, filter components and pattern comparison operation in accordance with the present disclosure.

(14) FIG. 14 illustrates the detection of a micro-crack in a gear shaft.

DETAIL DESCRIPTION OF DISCLOSURE

(15) An operator network 5 according to the invention is shown in FIG. 2 in an exemplary configuration. It comprises a decision operator 6, which here is connected on the input side to a fault pattern operator 7, an event pattern operator 8 and/or a free pattern operator 9 and on the output side to a signal operator 10, a counting operator 11 and/or a memory operator 12.

(16) Each operator is a program module with predefined input and/or output interfaces and internal logic, for filtering data, collecting data, processing data, making decisions, storing data, interacting with computers, databases, users and/or the like, and/or activating devices such as signals, motors, shutdown devices and/or the like, etc.

(17) The fault pattern operator 7 shown here can be constructed as shown in FIG. 3. A data stream 13, for example as a raw data stream of a sensor, in particular a sound signal sensor, or as a data stream processed by in particular other operators, for example as a frequency-filtered and/or smoothed data stream, is passed to a position determination operator 14 which, taking into account a process time provided by a time operator 15, to determine the temporal position of the data within the process and feeds it to a pattern comparison operator 16. An energy determination operator 17 is also connected to the operators 13, 15, to determine the energy in the data stream, for example in the form of the sound intensity, and correlates this with the process time in order to feed the result to the pattern comparison operator 16. Using these two input data streams and, if necessary, as shown, the data stream from the data stream operator 13, the time component from the time operator 15 and/or pattern data from a pattern operator 18, which is advantageously connected to a pattern database or can otherwise retrieve patterns for events, the pattern comparison operator 16 determines a measure of the agreement between the data supplied by the data stream operator 13 and a pattern supplied by the pattern operator 18, or a number of measures for a number of patterns. For example, the measure may be a number between 0% and 100% or between 0 and 255 or the like. The level of agreement is provided at the result output 19.

(18) The pattern operator 18 accesses fault patterns as occur during the process, such as the impact of the punch on the component to be straightened, a humming noise due to vibrations of the straightening device, ambient influences such as knocking by operating personnel, grinding noises of the tool on the workpiece, etc. At the result output 19, for example, a probability of 67% can be output for the identification of a fault pattern.

(19) The event pattern operator 8 can be structured in substantially the same way as the fault pattern operator 7. Thus, its logic may also include a data current operator, a time operator, a position determination operator, an energy determination operator, a pattern comparison operator and/or a pattern operator, which may be similarly or identically connected and internally configured with weighting factors and the like, the pattern operator accessing event patterns from a pattern library representing faults, such as patterns for stress cracks or tool breakage during machining and the like. The event pattern operator 8 provides, at its result output 19, see FIG. 2, a value indicating that such an event has occurred, for example 82%.

(20) As shown in FIG. 4, the free pattern operator 9 can have, in addition to a data stream operator 13, a time operator 15 and a position determination operator 14 as shown in FIG. 3, an energy determination operator 20, which calculates, for example, the energy, such as the sound signal intensity, and/or duration, and provides it at its result output 19. No pattern recognition takes place here, but instead the analysis of the energy.

(21) The processes and/or results can be displayed on a screen or the like by means of visualization operators 21.

(22) According to the invention, the operators can be arranged and configured in a graphical user interface. Thus, they can be dragged and dropped with the mouse or the like from a list of operators and arranged on a worksheet. Data connections between the inputs and outputs illustrated as lines in FIGS. 2, 3 and 4 are also graphically configurable. By right clicking or the like on an operator, its internal parameters or logic can be selected and/or changed. For example, in the energy determination operator 20, the formula for calculating the energy value can be determined, in the pattern comparison operator 16, the algorithm can be selected to determine the level of agreement between the acquired data and a pattern (for example by adding up the differences or the like), etc.

(23) The graphical user interface enables the user to implement, adapt and maintain an operator network individually tailored to the particular process, quickly and flexibly without having to rely on the manufacturer or on programmers.

(24) The decision operator 6 is of particular importance in the configuration insofar as it enables discrimination in a reliable manner for components/workpieces that are free of faults (OK) or faulty (not OK, NOK).

(25) In the example shown in FIG. 2, OK, NOK and NO inputs 22, 23, 24 of the decision operator 6 are connected to the result outputs 19 of the operators 7, 8, and 9.

(26) At the OK input 22, the fault pattern operator 7 provides a (probability) value together with further information that an insignificant interfering signal is currently being detected, i.e. the workpiece is OK. At the NOK input 23, the event pattern operator 8 provides a (probability) value and further information that a significant event such as a stress crack is currently being detected, i.e. the workpiece is not OK (NOK). The NO input 24 provides further information without pattern comparison (NO pattern recognition). The further information can include an energy value, such as the sound intensity and/or a duration, such as the time duration of the energy from exceeding a threshold value until it falls below the threshold value, or similar.

(27) In accordance with the invention, the decision operator 6 is configured to contain multi-dimensional decision fields in cascading form and, in the basis of the result of the cascaded multi-dimensional decision, sends a trigger signal to the signal operator 10 to mark a workpiece as faulty (or possibly free of faults) so that a faulty workpiece can be rejected. Likewise, the results and/or data can be stored by passing them on to the memory operator 12, and the rejection per time unit or in total can be recorded via the counting operator 11 and displayed at the visualisation operator 21, for example in the form of a scale, in order to visualize trends.

(28) The first multi-dimensional decision field is illustrated in FIG. 5. It compares the OK value from the fault operator 7 and the NOK value from the event pattern operator 8. In the above example, an OK value of 67% and an NOK value of 82% have been determined for stress phase 1 in FIG. 1. This corresponds to data point 25, which is above a boundary line 26 configurable via the graphical user interface. A further data point 27 is below the boundary line 26. Data points 25, 26 are ambiguous assignments, i.e. the events could very likely be either an insignificant interfering signal or a significant event pattern signal, while further data points 28, 29 are less ambiguous because they are either 0% NOK or 0% OK.

(29) The boundary line 26 serves here as the first step of discrimination. For the next stage, the ambiguous data points are fed into a multi-dimensional NOK decision field according to FIG. 6, a multi-dimensional OK decision field according to FIG. 7, and/or a multi-dimensional NO decision field according to FIG. 8.]

(30) Data point 25 is thus plotted with its further data, here, by way of example, in relation to energy and duration, and is located above an NOK boundary line 30 in FIG. 6, below an OK boundary line 31 in FIG. 7, and outside the hatched NOK range 32 between the NO boundary lines 33, 34 in FIG. 8. In further embodiments, a decision field of two or more dimensions can include the parameters energy, duration, temporal position, counter, temporal scaling, amplitude scaling and the like and/or, instead of a boundary line, a boundary surface can be used.

(31) The decision operator 6 is expediently configured to evaluate the data point 25 as NOK, since the data point 25 is in the NOK range 32 with regard to any two of the three operators OK, NOK and NO.

(32) The boundary lines 26, 30, 31, 33, 34 can be individually adapted to the particular process.

(33) Another dimension in FIGS. 5, 6, 7 and 8 can be time. It is therefore expedient to set the boundary lines 26, 30, 31, 33, 34 depending on the time in the process. For example, at the beginning of each stress phase 1, 3 the force exerted on the workpiece during bending and straightening and thus the probability of breakage is low, whereas at the end of the stress phases 1, 3 it is maximum. Thus, it may be expedient to set, at the beginning of the stress phase, a boundary line 26 which is more generous than the boundary line 26, while at the end of the stress phase a boundary line 26 is set, at which an NOK event is assumed earlier due to the high force acting (see FIG. 5). A data point 27 can therefore be OK if it is recorded at the beginning of the stress phase, while at the end it is above the boundary line 26 and NOK. The temporal course between the boundary line 26 at t=start of stress phase at 0 ms over the boundary line 26 to the boundary line 26 at t=end of stress phase at, for example, 1000 ms can be set arbitrarily, for example linearly.

(34) The above-mentioned operators can be cascaded and/or arranged in parallel, as desired.

(35) It may therefore be expedient to provide a further decision operator 6, see FIG. 9, which monitors the maintenance condition of the bending device. Here, patterns for events can thus be evaluated differently, for example a data point in FIG. 7 with high energy and long duration can indicate a problem with the machine and it may be provided to stop the machine.

(36) Further operators can be provided, for example a notification operator, which automatically sends an e-mail or other message with relevant information to the maintenance service and/or the person responsible for the process.

(37) In a further embodiment, a difference operator can be provided. The difference operator is configured in such a way that it identifies a known pattern, for example a more or less constant background noise or a recurring interfering noise or other fault data, adapts the associated pattern as precisely as possible to the magnitude of the data stream, and subtracts it if there is sufficient agreement. Then, a cleaned data stream remains as the difference, which can be fed to the other operators. A possible arrangement is illustrated in FIG. 10, in which a difference operator 35 is connected downstream of the fault pattern operator 7 and extracts fault patterns from the data stream 13 on the basis of the fault pattern identification and feeds the cleaned data stream 36 to further processing, for example as input data stream in FIG. 2 for the operators 7, 8, 9, wherein the subsequent operator 7if presentcan be set differently, or the fault pattern operator 7 only identifies certain fault patterns, such as ripples, scatter, etc.

(38) In the following, the operator model according to the invention is explained by means of further examples. In particular, it can be executed on distributed, heterogeneous and asynchronously working information technology hardware and systems.

(39) Analyses are carried out in the measuring instrument according to the invention by a network of operators.

(40) At the beginning, there is an operator that enables access to input data.

(41) A measurement data stream coming from an external source can sample sensor data over channels with up to 100 M samples and 16 bit or 4 Ms and 24 bit and can calculate in real-time an FFT with 25000 spectra/s per channel and provide both as input data stream.

(42) 32 digital IOs for machine control and Profibus and Profinet for machine communication can also be provided as inputs and outputs for the operators.

(43) The analysis sequences are created in the graphical operator editor according to the invention, see FIG. 11 as an example.

(44) At the top level, processors (clock slot) execute analysis programs (custom operators).

(45) Custom operators represent containers for subprograms, in which complex networks of further subprograms and prefabricated operators can be implemented; see FIG. 12, which shows an example of a network for pattern analysis and decision making.

(46) On the next level, a network is visible that combines data stream, filter components and pattern comparison operations; see FIG. 13.

(47) In the editor, the individual operators present more or less complex GUI components that define the connection to the data and parameterise the way the operators work.

(48) Via the connections, data is exchanged in the form of some basic types, Bool, Number, int64 etc., and more complex data for example in the form of JSON objects.

(49) JSON objects are also exchanged with the GUI for parameterisation.

(50) New operators can be integrated via plugin.

(51) A new data source must therefore include, above all, an operator plugin.

(52) In the real-time measurement process, the data are processed by the operators and, as shown in this example, the best fitting pattern is found, scaled, and its counterpart in the data stream is analyzed.

(53) At the same time, an evaluation is made according to the needs of the process, and a decision is made, for example, with regard to the quality of the workpiece or the performance of a production process.

(54) In FIG. 14, for example, a micro-crack in a gear shaft is detected. A 50 ms long section of structure-borne sound data from an industrial straightening machine for gear shafts is shown. The landscape is created on the basis of 25000 spectra/s after FFT of the microphone data, which in this case are evaluated up to approx. 800 kHz.

(55) All data can be stored on measuring devices in RAW format and in various compressions and can be analysed both in real time as well as after process (for learning processes).

(56) Examples for sensors according to the invention are, among others, structure-borne sound microphones and magnetic sensors, as well as broadband analogue amplifiers (>100 MHz) for data acquisition in industrial processes.

(57) The operator concept can be extended to distributed, heterogeneous and asynchronously operating information technology hardware and systems. Individual sections of a large operator network can run on distributed systems.

(58) The sensors according to the invention can collect and pre-process data, i.e. they can be designed as smart sensors.

(59) A real-time processing system can perform spectral analysis and identify patterns. A PC makes complex decisions. The hardware and software used can be very different. As long as the interfaces of the operator system are supported, any device, for any service can be integrated into the overall processing.

(60) In addition to the actual device, the manufacturer of a device, a smart sensor, also provides a module for operator integration with a GUI for carrying out the parameterisation and making the inputs and outputs available to the other participants in the network.

(61) The data streams can be both analogue and digital and can be transported on different carriers and also support different protocols, their respective operators providing access to the data in compatible form.

(62) In all embodiments, learning phases can be provided in which the sensors can transmit large amounts of data; learning is slow. Analysis networks can be created and parameterised on the basis of the learning data. Parts of the operations are then carried out in the sensor itself, data is already pre-processed and evaluated and, in the best case, results are already transmitted.

(63) Systems with wireless power supply and wireless communication can be completely sealed and thus built up in a process-safe way, their use can be maintenance-free.

(64) For network integration, an OS (for example Linux system) with TCP/IP connectivity can run on each sensor. Alternatively, a sensor hub can be provided, which contains the OS system and brings the sensor into the network. The forwarding can be wireless or wired.

(65) For the supply of wireless components such as sensors, in accordance with the invention methods of energy harvesting and wireless energy transmission are preferably used in addition to battery supply.

(66) If the data are mainly provided by the sensor system via TCP/IP, an FPGA-based smart-switch, for example, can advantageously represent a collection point and further processing stage. This smart-switch carries out parallel computing on the connected data streams and then makes its results available again via TCP/IP or also via PCIe, USB, etc.

LIST OF REFERENCE SIGNS

(67) 1 Stress phase 2 Stress phase 3 Relief phase 4 Landscape 5 Operator network 6, 6 Decision operator 7 Fault pattern operator (Pattern recognition operator) 8 Event pattern operator (Pattern recognition operator) 9 Free pattern operator 10 Signal operator 11 Counting operator 12 Memory operator 13 Data stream operator 14 Position determination operator 15 Time operator 16 Pattern comparison operator 17 Energy determination operator 18 Pattern Operator 19 Result output 20 Energy determination operator 21 Visualisation operator 22 OK input 23 NOK input 24 NO input 25 Data point 26, 26, 26 Boundary line 27 Data point 28 Data point 29 Data point 30 NOK boundary line 31 OK boundary line 32 NOK range 33 NO boundary line 34 NO boundary line 35 Notification operator, Machine control operator and/or Difference operator 36 Data stream 37 Graphical user interface