INDUSTRIAL EXTRUDER, PROCESS AND METHOD THEREOF

Abstract

Proposed is an industrial smart extruder and extrusion method, the industrial, intelligent extruder acting as a conveying device that uniformly squeeze solid to viscous masses out of a forming orifice under high pressure and temperature according to the operating principle of the Archimedean screw. Material is processed by hot extrusion or cold extrusion or warm extrusion or friction extrusion or micro extrusion by the extruder, and the material extruded by the extruder at least includes food products or metals or polymers or ceramics or concrete, or modelling clay. The industrial, intelligent extruder includes a smart device with a ML- or AI-based core engine controlling and/or steering and/or optimizing the operation of the extruder automatically.

Claims

1. An extruder system, comprising: a feeder, an extruder, a shaping opening, a collector, and an extruder control, the feeder feeding plastically deformable and/or viscous input material to the extruder, the extruder continuously pressing the input material to and out of the shaping opening forming an output material as extrudate, wherein the collector collects the extrudate for further processing, the extrusion process is controllably steered during operation by the extruder control, the extruder control comprises programmable logic to set and adapt operational setting parameter values of operational units of the feeder, the extruder, the shaping opening, and the collector, the extruder control further comprises a digital controller to signal and steer the programmable logic, for steering the programmable logic, the digital controller captures input parameter values at least comprising process parameter values and/or operational setting parameter values and/or material characteristics parameters values and/or environmental measuring parameter values, the input parameter values comprise sensory parameter values measured by sensors with the feeder and/or the extruder and/or the shaping opening and/or the collector, the extruder control includes a repository storage with an adaptive, digital database holding a plurality of selectable, structured data records for storing digital recipes, each of the selectable data record at least comprising material characteristics parameters of the input material and target material characteristics parameters of the extrudate and initial operational setting parameters giving an initial setting of operational setting parameters for the operational of the extruder system, the input parameter values further include parameter values of a selected data record, the digital controller includes a machine learner to monitor and classify value patterns of the input parameter values and adapting the operational setting parameter values of operational units of the feeder and/or the extruder and/or the shaping opening and/or the collector to align measured material characteristics parameters values of the extrudate within the predefined tolerance ranges, wherein the machine learner at least comprises at least a Deep Learning (DL) structure comprising one or more a plurality of Neural Network (NN) structures and/or one or more statistical modelling structures providing output parameter values based on the input parameter values indicating parameter values adaptions required to align the measured material characteristics parameters values of the extrudate within the predefined tolerance ranges, and the machine-learning based Deep Learning (DL) structure at least comprise a cascade of multiple layers of nonlinear processors for feature extraction and signal transformation, each successive layer using the output of the previous layer as an input providing supervised learning at least for classification and/or unsupervised learning at least for pattern recognition, and the extruder system includes digital signaling to steer the programmable logic and associated operational units via the digital controller to controllably and steered extrude an extrudate having material characteristics parameters values within a predefined tolerance range of predefined target parameter values, the extrusion process is autonomously adapted by the machine-learning unit by automatically adapting the operational setting parameter values of operational units of the feeder and/or the extruder and/or the shaping opening and/or the collector to align measured material characteristics parameters values of the extrudate within the predefined tolerance ranges by time-based monitoring of the input measuring parameter values.

2. The extruder system according to claim 1, wherein the adaptive, digital database is a digital library, the repository storage includes a network interface to provide access via a data transmission network to the structured data records for selection and/or adaption and/or generation of the structured data records.

3. The extruder system according to claim 1, wherein failures during the extrusion process are automatically detected by the machine learner based on the measured and monitored input parameter values, wherein an alert signaling and/or steering signaling is generated upon detection of a predicted failure within the extrusion process.

4. The extruder system according to claim 1, wherein the DL structure at least comprises a Convolutional Neural Network (CNN) structure as deep neural network.

5. The extruder system according to claim 4, wherein measured and/or captured input parameter value pattern are classified and selected by convolutional layers and pooling layers of the CNN structure, the pooling layers reducing a dimension of a feature map of the measured and/or captured input parameter value pattern and reducing processing complexity within the machine learner to adapt the operational setting parameter values of operational units.

6. The extruder system according to claim 1, wherein the repository storage further comprises an operational data storage to store historical operation data, historical operation data comprising historical input parameter values, historical material characteristics parameter values and predefined target parameter values, the machine learner being trained by applying historical operation data.

7. The extruder system according to claim 1, wherein the material characteristic parameters include texture and/or density and/or color and/or anisotropy and/or chemical composition and/or thickness and/or degree of polymerization and/or moisture content and/or protein content and/or starch content and/or fiber content and/or particle size and/or surface structure and/or tolerance range.

8. The extruder system according to claim 1, wherein the operational setting parameter comprises a screw speed of a screw of the extruder pressing the input material and/or addition rate by the feeder of at least one ingredient for composing the input material and/or conditioning setting of a conditioner of the extruder for cooling or heating the input material in the extruder and/or conditioner of the shaping opening and/or positioning size for the shaping opening area.

9. The extruder system according to claim 1, wherein the target parameters comprise at least one of the material characteristic parameters, and/or process parameters at least comprising energy consumption of the extruder system.

10. The extruder system according to claim 1, wherein material parameter values of the input material and/or an ingredient of the input material are automatically determined by the machine learner adapting a dosing process of the feeder by adapting operational setting parameters of the feeder.

11. A decentralized extruder networked system comprising: two or more extruder systems according to claim 1 and a central digital controller comprising a central repository having at least one adaptive central digital database containing structured data sets for storing digital recipes and/or ingredients and/or products, wherein at least one of the digital controller of the plurality of extruder systems is to be given read and/or write access to the structured central data records via the data transmission network for selecting and/or adapting and/or generating the structured central data records, and the central data records generated by an extruder system have at least one data classification parameter charactering this generating extruder system or factors influencing the generating extruder system.

12. The decentralized extruder networked system according to claim 11, wherein the data classification parameter, comprises the country of operation of the generating extruder system and/or the operator of the generating extruder system and/or the generating extruder system identification and/or the extruder type, in order to classify the centrally stored data records in the at least one central database with data classification parameter.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0093] The present invention will be explained in more detail by way of example in reference to the drawings in which:

[0094] FIG. 1 shows a diagram illustrating schematically an exemplary simple schematic of extrusion processing, showing the transformation of raw ingredient (such as grain flour or starch) to finished product.

[0095] FIG. 2 shows a block diagram illustrating schematically an exemplary flow chart of an extrusion processing production line. The extrusion process begins with characterizing and receiving the ingredients. The ingredients used are crucial to the extrudate consistency at the end of the processing line. The ingredients then undergo mixing and/or preconditioning, which can be done with an equipment such as ribbon blenders and preconditioners to ensure uniformity as they enter the extruder. However, mixing and preconditioning is optional for certain products. Extrusion processing then follows, which is the main cooking step where the raw ingredients are transformed into the cooked and formed products. Post-extrusion processing operations, such as cutting the extruded products into appropriate sizes, drying the products to the desired moisture, as well as seasoning or coating to provide the desired flavor and taste to the products before they go on to packaging. Along with these major sets of processing, Convolutional Neural Networks CNN their intended uses.

[0096] FIG. 3 shows a block diagram illustrating schematically an exemplary extrusion system as a multiple input and multiple output (MIMO) system. FIG. 3 demonstrates exemplary the various inputs and outputs associated with the inventive extruder system.

[0097] FIG. 4 shows a diagram illustrating schematically an exemplary of a digital controller comprising an interface to the programmable logic.

[0098] FIGS. 5 and 6 provides a description an input parameter and a digital recipe.

[0099] FIG. 7 shows a diagram illustrating schematically exemplary of a neural network.

[0100] FIG. 8 shows a diagram illustrating schematically exemplary a Convolutional Neural Networks comprising a feature extraction layer and a classification layer.

[0101] FIG. 9 shows a diagram illustrating schematically exemplary an extrusion network comprising several extruder systems communicating via a data transmission network.

[0102] FIG. 10 shows a diagram illustrating schematically an exemplary simple schematic of extrusion processing, showing the transformation of raw ingredient (such as grain flour or starch) to finished product.

[0103] FIG. 11 shows a block diagram illustrating schematically an exemplary flow chart of an extrusion processing production line. The extrusion process begins with characterizing and receiving the raw ingredients. The raw ingredients used are crucial to the product consistency at the end of the processing line. The raw ingredients then undergo mixing and/or preconditioning, which can be done with the equipment such as ribbon blenders and preconditioners to ensure uniformity as they enter the extruder. However, mixing and preconditioning is optional for certain products. Extrusion processing then follows, which is the main cooking step where the raw ingredients are transformed into the cooked and formed products. Post-extrusion processing operations, such as cutting the extruded products into appropriate sizes, drying the products to the desired moisture, as well as seasoning or coating to provide the desired flavor and taste to the products before they go on to packaging. Along with these major sets of processing, there may be additional steps depending on the type of products being produced and their intended uses.

[0104] FIG. 12 shows a block diagram illustrating schematically an exemplary extrusion process as a multiple input and multiple output (MIMO) system. FIG. 3 demonstrates exemplary the various inputs and outputs associated with the inventive extruder system. The various extrusion processing parameters can be broadly classified into three categories: (1) independent parameters (input parameters), (2) system parameters (dependent parameters) and (3) product properties (output parameters).

[0105] FIGS. 13a/b show a diagram illustrating schematically an exemplary (a) thermoplastic single-screw extruder, (b) thermoplastic single-screw.

[0106] FIG. 14 provides a description of the screw terminology using a single-screw extruder screw elements.

[0107] FIG. 15 shows a diagram illustrating schematically exemplary corotating and counterrotating intermeshing screws.

[0108] FIG. 16 shows a diagram illustrating schematically exemplary intermeshing corotating double flight screws and a 45 kneading block.

[0109] FIG. 17 shows a diagram illustrating schematically exemplary corotating screws showing open channel in the intermeshing region.

[0110] FIG. 18 shows a diagram illustrating schematically exemplary general flow patterns in corotating and counterrotating twin-screw extruders (two flight screws in both cases).

[0111] FIG. 19 shows a diagram illustrating schematically an exemplary shear stress profile in counterrotating and corotating twin-screw extruders.

[0112] FIG. 20 shows a diagram illustrating schematically an exemplary possible arrangement of a corotating twin-screw extruder for a multi-process food extrusion system.

[0113] FIG. 21 shows a diagram illustrating schematically an exemplary extrudate swelling at the die exit.

[0114] FIG. 22 shows a diagram illustrating schematically an exemplary product expansion after exiting the die due to vapor pressure difference.

[0115] FIG. 23 shows a diagram illustrating schematically an exemplary intermeshing orifice plugs (discs) for severe screw restriction and dispersive mixing.

[0116] FIG. 24 shows a diagram illustrating schematically an exemplary twin-screw extruder showing the simulated pressure and temperature profiles within the extruder.

[0117] FIG. 25 shows a diagram illustrating schematically an exemplary screw design with two venting ports for degassing of volatiles.

[0118] FIG. 26 shows a diagram illustrating schematically an exemplary vent port adaptor design.

[0119] FIG. 27 shows a diagram illustrating schematically an exemplary extrusion process realized by the typical continuous thermo-mechanical process technology combining several unit operations like conveying, mixing, shearing, plasticization, melting, cooking, and polymerization. Those numerous unit operations lead to complex correlations between several variable parameters and their respective process response and thus, determining the product quality.

[0120] FIG. 28 shows a diagram illustrating schematically an exemplary AI-based or ML-based automated recipe management and optimization in an inventive extrusion process and inventive extrusion system.

[0121] FIG. 29 shows a diagram illustrating schematically an exemplary smart recipe selection, where smart recipe selection is the process of selecting a fitting recipe out of a set of predefined recipes based upon given descriptors of a product. The descriptor is defined as set of different characteristics of a product. Example characteristics would be color, fibrosity etc. Such sets would be nameable (in the context of high moisture extrusion) like Chicken or Fish etc.

[0122] FIG. 30 shows a diagram illustrating schematically an exemplary RDB, where RDB is a database maintained by the inventive system that can be used by the Operator to search/download new recipes. The database acts as storage for the automatic recipe selection, which validates and updates the recipe data by inserting insights of process runs that were recorded. It is based on ingredients characteristics such as color, fibrosity etc. and maps to achievable final product characteristics.

[0123] FIG. 31 shows a diagram illustrating schematically an exemplary IRPC system yielding data on raw materials used for the extrusion process. Its target is to act as the following: (a) an input data for the automatic recipe selection by mapping types of raw ingredients based on possible achievable product characteristics to the wished product of the customer/operator. It acts as an advisor: The customer/operator can either choose a specific recipe and the IIC will return whether with the given ingredient, the wished final product is feasible, or itself give a proposal on which ingredients should be used for the final product to receive the best results, and (b) An input for the smart process optimization system through predicted/anticipated raw product data.

[0124] FIG. 32 shows a diagram illustrating schematically an exemplary ingredients database of the inventive system, contains entries to raw product characteristics such as moisture, protein content etc.; process target information such as CO.sub.2 relevance etc. that can be opposed by measured outcomes of the final product classification that uses product characteristics as classification.

[0125] FIG. 33 shows a diagram illustrating schematically an exemplary smart process optimization which is the act of adjusting parameters of the extrusion process during production based on given targets, where targets are process targets (PT) as well as product targets (given by the recipe). The optimization process has to fulfill or exceed the targets given by the operator. The process optimizer is restricted in his optimization by deviation of product target, where the target is either a machine recipe or product characteristics. Process targets (PT) come in form of pre-engineered chains of commands.

[0126] FIG. 34 shows a diagram illustrating schematically an exemplary intelligent final product classifier which can e.g. use sensorics' information retrieved from the final product out of the extruder to map the product to final product characteristics. As information sources, the final product classifier can e.g. use data from the process itself (Temperatures, Pressures, SME etc.), as well as online final product measurements. To verify its measurements and to train the system, offline final product measurements can e.g. be used.

[0127] FIG. 35 shows a diagram illustrating schematically exemplary process targets. Process targets are sets of process commands that affect how the process is run. Possible process targets are (i) minimizing energy consumption of the process, (ii) maximizing the throughput of the process, and (iii) minimizing CO.sub.2 emissions.

[0128] FIG. 36 shows a diagram illustrating schematically an exemplary automated rework process. Smart rework ingestion is the process of using non-usable product coming out of the extrusion process again by feeding the product back into the process, where the amount of product to be reused can be varied. This service gathers its input data from the following sources: The extrusion process itself, as well as the final product classification, that yields information on the products state.

[0129] FIG. 37 shows a diagram illustrating schematically an exemplary full overview with structures required for the inventive automated recipe management and optimization.

[0130] FIG. 38 shows a diagram illustrating schematically an exemplary CO.sub.2e monitoring dashboard according to an embodiment variant of the inventive system.

[0131] FIG. 39 shows a diagram illustrating schematically an exemplary pre-conditioning process of an exemplary extrusion process.

[0132] FIG. 40 shows a diagram illustrating schematically an exemplary extrusion process of an exemplary extrusion process. Pre-conditioning (see FIG. 39) and extrusion (see FIG. 31) form the basics of the whole extrusion process.

DETAILED DESCRIPTION OF THE INVENTION

[0133] FIGS. 1 to 8 illustrate, schematically, an architecture for a possible implementation of an embodiment of the inventive industrial extruder system 1 and extrusion process. The industrial extruder system 1 comprises a feeder 101, an extruder 102, a shaping opening 103 (die), collection means 104, and an extruder control 500, the feeder 101 feeding plastically deformable and/or viscous input material 301 to the extruder 102, the extruder 102 continuously pressing the input material (301) to and out of the shaping opening (103) forming an output material as extrudate 300 according to the operating principle of the Archimedean screw. Input material 301 is processed by hot extrusion or cold extrusion or warm extrusion or friction extrusion or micro extrusion by the extruder 1. The input material 301 extruded by the extruder system 1 can at least comprise food products or metals or polymers or ceramics or concrete or modelling clay. The collection means 104 collects the extrudate 300 for further processing, for example a post extrusion processing comprising cutting and/or drying and/or seasoning and/or frying and/or coating, according to FIG. 2.

[0134] The extrusion process is controllably steered during operation by the extruder control 500, and wherein the extruder control 500 comprises programmable logic 107 for setting and adapting operational setting parameter values 208 of operational units 105 of the feeder 101, the extruder 102, the shaping opening (die) 103, and the collection means 104. The programmable logic 107 comprises programmable logic controller of operational units, for example a programmable logic controller of a frequency converter for controlling the motor 108. The operational setting parameter 201 comprises a screw speed of a screw of the extruder pressing the input material 301 and/or addition rate by the feeder of at least one ingredient for composing the input material 301 and/or conditioning setting of the conditioner unit 122 of the extruder 102 for cooling or heating the input material 301 in the extruder 102 and/or conditioner of the shaping opening 103 and/or positioning size for the shaping opening 103 area.

[0135] In that the extruder control 500 further comprises a digital controller 502 for signaling and steering the programmable logic 501, wherein for steering the programmable logic 501, the digital controller 502 captures input parameter values 200 at least comprising process parameter values/or and operational setting parameter values (208) and/or material characteristics parameter values 202 and/or environmental measuring parameter values 209. The material characteristic parameter 202 comprises texture and/or density and/or color and/or anisotropy and/or chemical composition and/or thickness and/or degree of polymerization and/or moisture content and/or protein content and/or starch content and/or fiber content and/or particle size and/or surface structure and/or tolerance range. Further the input parameter values 200 comprise sensory parameter values 207 measured by sensors 111 with the feeder 101 and/or the extruder 102 and/or the shaping opening 103 and/or the collection means 104 as illustrate in FIG. 3.

[0136] The extruder control 500 comprises a repository storage unit 503 with an adaptive, digital database 504 holding a plurality of selectable, structured data records 507 for storing digital recipes, each of the selectable data record 507 at least comprising material characteristics parameters 202 of the input material 301 and target material characteristics parameters 204 of the extrudate 300 and/or initial operational setting parameters giving an initial setting of operational setting parameters 201 for the operational 105 of the extruder system 1, wherein the input parameter values 200 further comprise parameter values of a selected data record 507 as illustrated FIG. 6. The extruder system 1 comprises a digital signaling for steering the programmable logic 501 and associated operational units 507 by means of the digital controller 502 to controllably and steered extrude an extrudate 300 having material characteristics parameters values 202 within a predefined tolerance range of predefined target parameter values 206 as illustrated in FIG. 3. 2. The digital database 504 is realized as a digital library, the repository storage unit 503 having a network interface 506 providing access via a data transmission network 505 to the structured data records 507 for selection and/or adaption and/or generation of the structured data records 507. Data transmission is the transfer and reception of data in the form of a digital bitstream or a digitized analog signal transmitted over a point-to-point or point-to-multipoint communication channel. Examples of such channels are copper wires, optical fibers, wireless communication using radio spectrum, storage media and computer buses.

[0137] Furthermore, the digital controller 502 comprises a machine learning unit 510 monitoring and classifying input parameter values 200 patterns and adapting the operational setting parameter values 208 of operational units 105 of the feeder 101 and/or the extruder 102 and/or the shaping opening 103 and/or the collection means 104 to align measured material characteristics parameters values 202 of the extrudate 300 within the predefined tolerance ranges. The machine learning unit 510 can e.g. realized by comprising a Deep Learning (DL) structure e.g. comprising one or more Neural Network (NN) structures as illustrated in FIG. 7. The one or more Neural Network (NN) structures can e.g. comprise an input layer 601, at least one hidden layer 602 and an output layer 603; and/or one or more statistical modelling structures providing output parameter values 213 based on the input parameter values 200 indicating parameter values adaptions required to align the measured material characteristics parameters values 202 of the extrudate 300 within the predefined tolerance ranges. The machine-learning based DL structure at least comprise a cascade of multiple layers of nonlinear processing units 512 for feature extraction 512a and signal transformation 512b, and wherein each successive layer uses the output of the previous layer as an input providing supervised learning at least for classification and/or unsupervised learning at least for pattern recognition. Supervised learning be separated into two types of problems when data mining: classification and regression: (i) Classification problems use an algorithm to accurately assign test data into specific categories. Or, in the real world, supervised learning algorithms can be used to classify spam in a separate folder from your inbox. Linear classifiers, support vector machines, decision trees and random forest are all common types of classification algorithms. Preferable the DL structure at least comprises a Convolutional Neural Network (CNN) as illustrated in FIG. 8, structure as deep neural network.

[0138] Regression is another type of supervised learning method that uses an algorithm to understand the relationship between dependent and independent variables. Regression models are helpful for predicting numerical values based on different data points, such as sales revenue projections for a given business. Some popular regression algorithms are linear regression, logistic regression, and polynomial regression. Unsupervised learning uses machine learning algorithms to analyze and cluster unlabeled data sets. These algorithms discover hidden patterns in data without the need for human intervention hence, they are unsupervised). Unsupervised learning models are used for three main tasks: clustering, association, and dimensionality reduction: (i) Clustering is a data mining technique for grouping unlabeled data based on their similarities or differences. For example, K-means clustering algorithms assign similar data points into groups, where the K value represents the size of the grouping and granularity. This technique is helpful for market segmentation, image compression, etc. (ii) Association is another type of unsupervised learning method that uses different rules to find relationships between variables in a given dataset. These methods are frequently used for market basket analysis and recommendation engines, along the lines of Customers Who Bought This Item Also Bought recommendations; (iii) Dimensionality reduction is a learning technique used when the number of features or dimensions) in a given dataset is too high. It reduces the number of data inputs to a manageable size while also preserving the data integrity. Often, this technique is used in the preprocessing data stage, such as when autoencoders remove noise from visual data to improve picture quality. The measured and/or captured input parameter value 200 pattern are classified and selected by means of convolutional layers 612 and pooling layers 613 of the CNN structure, the pooling layers 613 reducing the dimension of a feature map of the measured and/or captured input parameter value (200) pattern and thus reducing the processing complexity within the machine learning unit 510 to adapt the operational setting parameter values 201 of operational units 105.

[0139] Failures during the extrusion process are automatically detected by the machine-learning unit based on the measured and monitored input parameter values 200, wherein an alert signaling and/or steering signaling is generated upon detection of a predicted failure within the extrusion process. Preferable the digital controller 502 comprises a failures handling unit generating based on alert signaling and/or steering signaling an error signal. Operating setting parameter comprising the error signal for steering program logic 501, in particular, for setting an error function of the respective program logic controller of the respective operational unit 105. The failure handling unit is configurable for a given alert signaling and/or steering signal. Preferable the digital controller comprises a human machine interface for signaling alert signaling.

[0140] An extruder network 2 comprising two or more extruder systems 1 and a central digital controller 514 comprising a central repository 512 having at least one adaptive central digital database 513 containing structured data sets 515 for storing digital recipes and/or ingredients and/or products, and wherein at least one of the digital controller of the plurality of extruder systems 1 is to be given read and/or write access to the structured central data records 515 via the data transmission network 505 for selecting and/or adapting and/or generating the structured central data records 515, and wherein the central data records generated by an extruder system 1 have at least one data classification parameter 211 charactering this generating extruder system or factors influencing the generating extruder system 1. the data classification parameter 211, comprises the country of operation of the generating extruder system and/or the operator of the generating extruder system 1 and/or the generating extruder system 1 identification and/or the extruder type, in order to classify the centrally stored data records in the at least one central database with data classification.

I. Uniform Product Quality & Man Less Operation

[0141] Extrusion is a continuous thermo-mechanical process technology combining several unit operations like conveying, mixing, shearing, plasticization, melting, cooking, and polymerization. Those numerous unit operations lead to complex correlations between several variable parameters and their respective process response and thus, determining the product quality (see FIG. 18). The current situation requires time- and cost-intense offline analysis, well-educated operators, and lab personnel to evaluate product quality and to adjust the process and thus product quality accordingly. Due to the nature of offline analyses, process adjustments are done time-delayed only. Hence, online measurements to evaluate product quality in real-time would allow immediate process adjustments and consistent product quality. Further, offline analysis would be reduced lowering the expenses on both, analysis, and personnel.

[0142] The set-up of an extruder and its periphery accounts for the fixed parameters being screw configuration, barrel length, cooling die geometry, and cooling die inserts. Variable parameters are temperature, screw speed, moisture addition rate, solid addition rate, oil addition rate, and locations of liquid addition like water, oil, flavors, mineral solution, acids, caustics. Besides that, the introduction of nitrogen (N2) as part of aeration technology is a variable parameter as well. The variable parameters determine the process response like specific mechanical energy (SME), thermal stress (STE), pressure at endplate, temperature at endplate, residence time (distribution), weighted average total strain (WATS, Forte), and flow profile/velocity at cooling die exit. All the previous mentioned process responses are quantities that are measurable. The SME is a calculated quantity from power consumption/torque measurements recorded from the main drive of the extruder, STE is calculated based on inlet and outlet temperatures, and volume flows of the heating cycles, pressure and temperature at the endplate are measured by pressure and temperature probes, flow profile and velocity is accessible by HD-cameras and underlying video analysis programming, residence time (distribution) is measured by using an inert indicator, e.g. colorant which is then also recorded by HD-cameras and tracked over time, the WATS (weighted average total strain) is a quantity calculated using fixed parameters (screw configuration, barrel length, cooling die geometry, and inserts), the screw speed and the residence time. As an embodiment variant, the ML-based or AI-based core engine 11 comprises an AI-controlled loop to alter the process response by adapting the variable parameters. Further, complex interactions among the process response and variable parameters with the product quality of the extrudates can be established. Since a change of one single variable parameter can have more than one effect on the process response, and thus affecting the product quality without clear correlation, Artificial intelligence, as realized by the invention, is a powerful tool to unveil these interactions for both daily production and quality assurance, but also for R&D purposes. Exemplary for that, there are prior art systems intended to measure and study the effect of SME on final product's anisotropy, however, it turned out, that when increasing the screw speed to change SME, also the temperature at the endplate was increased. Hence, by prior art systems, it was not possible to correlate the observed effects on product quality clearly.

[0143] The product quality of wet extrudates is most described by its texture, anisotropy, density, chemical composition, color, and degree of polymerization. To set-up an artificially controlled loop for real-time adjustments of the variable parameters the product quality must be analyzed online. The following table (table 2 is a non-exhaustive list) is showing applicable methodologies to enable consistent product quality throughout production.

TABLE-US-00002 Product quality Possible online methodology Texture Cutting force (torque) of cutter at die outlet Anisotropy Ultrasonic transducers to measure phase velocity and attenuation coefficient. The phase velocity and attenuation are correlated with offline samples exhibiting different anisotropic qualities being analyzed by tensile testing or compression tests. Density Ultrasonic transducers to measure phase velocity and attenuation coefficient. The phase velocity and attenuation are correlated with offline density measurements e.g., water displacement method. Chemical composition Near infrared spectroscopy (NIR) Raman spectroscopy Color HD-Camera, image analysis algorithm Color spectrophotometers (e.g., L*a* b*, CIELAB) Degree of Ultrasonic transducer/Raman spectroscopy/NIR to polymerization measure specific signal in the product flow. Both techniques must be correlated to protein extractability offline to calibrate the sensors accordingly. Thickness Ultrasonic transducers

Table 2 Shows Examples of Product Quality and Applicable Online Methodology

[0144] If it is not possible to directly measure some of the above stated product qualities, mapping needs to be achieved between measurable process parameters such as temperatures, pressures, and SME, as well as the above stated methods towards the stated product quality indicators. As an embodiment variant, machine learning modelling structures are e.g. used to perform the appropriate mapping.

[0145] The resulting final product quality indicators can be used for a multitude of different applications inside the extrusion process, for example as validation data of AI applications used previously in the process chain, such as automated process optimization, to define process targets by operators instead of using conventional recipes where it then is possible to decide between good and bad product, ingredient databases where ingredients can be mapped to achievable products.

II. Automated Recipe Management & Optimization

[0146] Regarding automated recipe management and optimization, there are various technical problems which are solved by the inventive extruder and extrusion system: (i) In the prior art systems, it is typically unclear whether the maximum potential of any extrusion machine has been reached; (ii) In the prior art systems, operators tend to have less know how as well as customers tend to have less know how; (iii) In the prior art systems, optimizing a recipe is typically very time consuming. All happens in the head of the technologist, i.e. empirically; (iv) In the prior art systems, optimized product characteristics are typically not clear and need to be defined approached by an empirical approximation process; and (v) In the prior art system, recipe management and optimization is still needed to be embedded in the extrusion process, i.e. forming a part of the standard extrusion process. FIG. 28 shows a diagram illustrating schematically an exemplary full overview with structures applicable to achieve the inventive automated recipe management and optimization.

(a) Smart Recipe Selection

[0147] Smart recipe selection is the process of selecting a fitting recipe out of a set of predefined recipes based upon given descriptors of a product (see FIG. 20). The descriptor is defined as set of different characteristics of a product. Example characteristics would be color, fibrosity etc., Such sets would be nameable (in the context of high moisture extrusion) like Chicken or Fish etc., Based on sets, included characteristics can be altered by the operator after selection. For example: The operator changes the color of the final product.

[0148] The selection process of the recipe builds up on different data sources: [0149] Recipe database RDB. [0150] Intelligent ingredients classification IIC [0151] Intelligent final product classification IFPC (described in a later chapter) [0152] Process target PT (described in a later chapter).Math. [0153] Selection of the operator

Recipe Database (RDB)

[0154] The RDB is a database maintained by the inventive system that can be used by the Operator to search/download new recipes (cf. FIG. 21). The database acts as storage for the automatic recipe selection, which validates and updates the recipe data by inserting insights of process runs that were recorded. It is based on ingredients characteristics such as color, fibrosity etc. and maps to achievable final product characteristics.

Ingredients Classification (IIC)

[0155] The IRPC system yields data on raw materials used for the extrusion process. Its target is to act as the following: [0156] an input data for the automatic recipe selection by mapping types of raw ingredients based on possible achievable product characteristics to the wished product of the customer/operator. It acts as an advisor: The customer/operator can either choose a specific recipe and the IIC will return whether with the given ingredient, the wished final product is feasible, or itself give a proposal on which ingredients should be used for the final product to receive the best results. [0157] An input for the smart process optimization system through predicted/anticipated raw product data.

[0158] It can be fed from the following sources: [0159] The ingredients database IDB (see below). [0160] online raw product analysis by inline devices such as NIR. [0161] offline raw product analysis such as a lab [0162] the final product classification IFPC that returns observed final product characteristics [0163] the process target selected by the operator [0164] an operator inserting raw product information manually

[0165] The IRPC uses measured online and offline raw product analytics data to detect deviations in the product. These deviations are stored in the raw ingredients database for training and validation of the IRPC model as well as its input data from the database. The IRPC sends its predicted raw product characteristics to the automatic process optimization, therefore taking influence into the process (see also FIG. 22).

Ingredients Database (IDB)

[0166] The ingredients database, maintained by the inventive system, contains entries to raw product characteristics such as moisture, protein content etc.; process target information such as CO.sub.2 relevance etc. that can be opposed by measured outcomes of the final product classification that uses product characteristics as classification.

Smart Process Optimization (SPO)

[0167] Smart process optimization is the act of adjusting parameters of the extrusion process during production based on given targets, where targets are process targets (PT) as well as product targets (given by the recipe). The optimization process has to fulfill or exceed the targets given by the operator. The process optimizer is restricted in his optimization by deviation of product target, where the target is either a machine recipe or product characteristics. Process targets (PT) come in form of pre-engineered chains of commands.

[0168] SPO uses the following data sources: [0169] The intelligent final product classifier (IFPC) returns the measured final product characteristics that leaves the process. [0170] The extrusion process yields live data from the process, such as product temperatures at given locations in the process chain, pressures of endplate and (cooling) dies, SME etc., These values act as deviation limit if recipe deviation is used. [0171] The smart recipe selector (SMR) sends final product characteristics to be achieved. [0172] The intelligent ingredients classifier (IIS) sends its predicted raw product characteristics information. [0173] The online/offline product measurement sends live/batch ingredient information [0174] The operator selects a process target (PT) [0175] The smart rework ingestion that inputs data on possible rework to be fed into the process.

[0176] The intelligent process optimizer can run in two different modes: Given that a smart recipe selector as well as the final product classifier is available, it can run in characterized product mode, where it controls the process based on achieved product character. Given that no smart recipe selector is available or no final product classifier is available, it could run in actual data only mode, where it would optimize based on inline measurement.

Intelligent Final Product Classification (IFPC)

[0177] The intelligent final product classifier uses sensorics' information retrieved from the final product out of the extruder to map the product to final product characteristics. As information sources, the final product classifier can e.g. use data from the process itself (Temperatures, Pressures, SME etc.), as well as online final product measurements. To verify its measurements and to train the system, offline final product measurements can e.g. be used.

[0178] Its output data, that reflects the characteristics of the final product, is reused by a multitude of systems: [0179] The smart process optimizer that uses this data to regulate the process. The measurement is used as quality metric where the target product characteristics are trying to be achieved. [0180] The smart rework ingestion process where the analysis of the not-yet-usable product is used to analyze how much of it can be fed back into the process right away to reduce waste [0181] The intelligent ingredients classifier that uses the output of the IFPC to map ingredients to achievable final products.

Process Targets (PT)

[0182] Process targets are sets of process commands that affect how the process is run (see FIG. 26). Possible process targets are: [0183] Minimizing energy consumption of the process [0184] Maximizing the throughput of the process [0185] Minimizing CO.sub.2 emissions

[0186] Process targets work as inputs for the following systems: [0187] Intelligent ingredients classifier, where they have an influence on its advisory function based on the target. For example, if the target CO.sub.2 emissions is chosen, ingredients will be selected based on their CO.sub.2 relevance. [0188] Smart recipe selector, where recipes are proposed and selected based on their fulfillment of specific targets [0189] Smart process optimizer, where command chains have an influence on the production process itself.

Automated Rework

[0190] Smart rework ingestion is the process of using non-usable product coming out of the extrusion process again by feeding the product back into the process, where the amount of product to be reused can be varied (see FIG. 27). This service gathers its input data from the following sources: The extrusion process itself, as well as the final product classification, that yields information on the products state. Whether product can be fed back into the extrusion process depends on the following parameters: [0191] Product characterization values set as target. Product that does not achieve the target can be reused. [0192] Product moisture. If the product is too moist, feeding back is not possible without the product sticking. [0193] Product chunk size. To big pieces will not be transportable.

[0194] Rework ingestion will have influence on smart process optimization. The product cannot be understood as normal ingredient of the process. Smart process optimization will receive a proposal of amount of rework to be ingested from the smart rework ingestion controller.

III. Ingredient Database (Cloud-Based) to Optimize Process Based on Historical Data

[0195] Extrusion allows to process a wide range of different ingredients. The ingredients and their characteristics are dependent on several environmental factors, but also on applied process steps during their production. Therefore, ingredients are prone to vary in their composition and functionality.

[0196] For high moisture extrusion of meat analog products, commonly different protein sources are used, being flours (protein about 50%), concentrates (protein 60-80%), and isolates (protein>80%), or even intermediate products like slurries from their production processes. In an embodiment variant, to enhance the intelligent extruder system, an ingredient database is created to optimize the extrusion process continuously. Besides different protein sources and water, formulations also include carbohydrates (i.e., starches, fibers, sugars), fat/oils, minerals, salts, acids, caustics, flavors. A comprehensive ingredient database must be compromised of compositional and functionality-related properties of the respective ingredient.

[0197] First, an ingredient is analyzed to determine its chemical composition being solid content (moisture content), protein content, fat content, carbohydrate content, and minerals.

[0198] Further, specific knowledge about protein fractions of an ingredient is of high importance as they determine the potential of the later texturization ability during the process. More specifically, the inter- and intramolecular interactions during texturization are a function of a specific amino acid profile. The most interactions among proteins upon high moisture extrusion are non-covalent interactions like hydrogen bonding, hydrophobic interactions, van der Waals interactions, and electrostatic interactions. In addition, sulfur-rich amino acids, in particular cysteine, form covalent disulfide bridges which are the dominant interactions to form thermo-stable high moisture extrudates. Consequently, a detailed analysis of the ingredients is required. Besides knowledge about the amino acid profile, advanced knowledge like denaturation temperature and kinetics, and aggregation temperature and polymerization kinetics are part of the ingredient database. Consequently, the ingredient database facilitates recipe development and gives insights whether an ingredient takes part in texturization or is acting as an inert filler in the final product.

[0199] Further analyses can be required to identify the specific carbohydrates to distinguish between sugars, fibers, and starches. For starch-rich ingredients it can be important to include knowledge about amylose-amylopectin ratio and gelatinization kinetics, and degradation. For sugar-rich ingredients it is of importance to include knowledge about melting temperatures, and caramelization.

[0200] In general, characteristics of the ingredients like true density, particles size distribution, flowability, wettability, dispersibility, water absorption, ionic strength and pH of slurries, viscosity, elastic modulus, phase transition from dispersion (viscoelastic liquid) into plasticized mass (viscoelastic solid), as well as time, temperature, and strain dependence of all these parameters are required for the ingredient database. The ingredient database is a continuous growing database which facilitates new ingredient evaluation and new recipe development. Further, optimized variable parameters for the extrusion process can be derived.

[0201] Considering the above list of required fluid and solid properties and the lack of detailed understanding of cooking, kneading, and conveying, it is unlikely that physics-based models of the full extrusion process will be available any time soon. However, notable attempts are currently being made on subsystems of the extrusion process, in particular the die flow of wet texturized proteins.

[0202] The quantification of many of the above-mentioned properties is a challenge on its own because laboratory devices often cannot mimic the real process, and online measurements are inaccurate since available sensors are either inadequate or unsuitably located. Often, only pressure and surface temperatures are available when viscosity and density values are needed to feed the material model.

[0203] Modelling viscoelastic flows in industrial processes with complex and interlocking geometries is exceedingly difficult and time consuming, but in principle feasible. Including phase transformation of complex fluids and semi-solids requires innovative simulation techniques, such as meshless methods. This is currently not the industry standard and an ongoing academic field of research. In the prior art, three software codes are known to have been used for similar challenges, or to be able to capture the main physics of such processes.

[0204] However, such simplified modelling approaches of the extrusion process could achieve only limited success. From today's perspective, creating material models and measuring corresponding material properties would require a vast amount of resources and time to get to the point where it is usable industrially. Therefore, data-based methods offer an alternative and more accessible way to (try to) relate input parameters with result quantities and build a model that would eventually allow to predict and optimize the process.

IV. Optimized & Fully Automated Startup Time

[0205] The today's situation for an extrusion process requires an individual start-up procedure by each operator resulting in a waste of raw materials and time (about 30 min), as start-up phase is evaluated objectively only. Hence, the inventive intelligent extruder system enables a more precise and optimized start-up phase when considering measured quantities to evaluate texture (e.g., cutting force at die) online and process responses (e.g. pressure at endplate). For example, the pressure probe at the endplate (transition from extruder barrel to cooling die) is a suitable quantity to rely on for automation of the start-up process. When the pressure at endplate rises above 2 bar, a filled screw section is indicated. Hence, in this embodiment variant, the production phase is initiated automatically.

[0206] In addition, online measurement (chemical composition e.g., by NIR) of the slurry during start phase enables its use as rework. For example, as an embodiment variant, the rework is injected into the barrels. In this case, the intelligent extruder system relies on the ingredient database to adjust variable parameters and finally preserve consistent product quality during production.

V. Safe Production

[0207] It is known in the prior art that plant-based meat analogues are more perishable and more susceptible to spoilage compared to meat due to high protein and moisture content as well as neutral pH, therefore more attention is needed in meat analogue production. In particular, an effective food safety management system is of great importance to ensure the safe production.

Online/inline CCP monitoring

Critical Control Point (CCP):

[0208] Temperature in the end plate

[0209] A temperature sensor is installed in end plate to record the temperature every 1 second. The real-time temperature curve will be displayed in a dashboard, which has following functionalities: [0210] Calculates the Log 5 reduction conditions based on the recipe [0211] Sets critical values (temperature, duration etc.) for key food safety parameters based on the Log 5 reduction conditions [0212] Alarm & warning functions when the temperature is below the critical threshold (CCP errors) [0213] Display a summary of alarms and a list of all time windows/time periods when the CCP errors occur [0214] Each alarm and CCP error can be selected for further analysis. [0215] Ability to trace the product to the errors and see what happened with the product as well as the actions taken (e.g. equipment cleaned and sanitized etc.) [0216] Generates reports showing CCP, critical limits, CCP errors, recipe, time frame, operator, corrective actions etc.

Online/Inline CCP Control.

[0217] When CCP errors occur and no corrective actions are taken from the operator, the control system will automatically adjust the process to increase the temperature to be above the critical threshold based on the recipe. [0218] A diversion flap is installed at the exist of the cooling die and the product which is produced in the time periods where CCP criteria is not reached will go to the waste bin

Verification of Machine Cleanliness

[0219] Optical sensors are installed in preconditioner (cf. FIG. 30), extruder (cf. FIG. 31), cooling die, conveyor, and cutter. After the cleaning, optical sensors are used to assess the cleanliness (by detecting residues and microorganism contamination etc.), An alarm system is set up to give warnings when the cleanliness does not meet the criteria and the corrective actions are needed. For the wet cleaning, visual check is normally not enough. A microbiological test should be conducted. An ATP test can be recommendable (rapid test: 5 min). A tick box is created in the dashboard, where the operator has to tick if a microbiological test has been carried out or not. Alternatively, the machine cleanliness can be determined by an indirect approach (e.g. microbiological test on the downstream product)

Safety Check Before Starting the Extrusion Process

[0220] Before starting the extrusion process, the control system will execute a self-check procedure: [0221] Checking the cleaning history (e.g. when was the last cleaning) [0222] Checking the microbiological test (e.g. when was the last test) [0223] Optical cleanliness check

[0224] Based on the feedback from the above-mentioned checks, the control system will decide if the extrusion process can be started, or further cleaning/test is required.

VI. Sustainability

[0225] CO.sub.2 equivalent (CO.sub.2e) is the standard unit for measuring carbon footprints and it is used as a common metric due to the fact that it takes all greenhouse gases into account and is readily available. It simplifies the problem of climate change and highlights carbon hotspots for targeted actions. Furthermore, it aligns to the ISO standards.

Ingredient Database for Consulting Service

[0226] For all relevant ingredients, their potential suppliers and associated product specification, cost and CO.sub.2e emission factor etc. are documented in the ingredient database. [0227] Based on the specific requirements of customers, suitable ingredient suppliers could be suggested. [0228] A summary of CO.sub.2e emission can be generated based on the selected ingredients, suppliers, and transportation etc. [0229] A comparison of CO.sub.2e emission of different ingredients, suppliers and transportation can be generated.

Co.SUB.2.e Monitoring of Process

[0230] When the system is connected to an expert insights system, the information on the ingredients, throughput, and energy consumption will be obtained, which can be used for the CO.sub.2e emission quantification of the whole process.

[0231] Following digital monitoring and expert services can be provided, e.g. using, inter alia, an appropriate CO.sub.2e monitoring dashboard (see FIG. 29): [0232] Real-time CO.sub.2e emissions of the whole process [0233] Breakdown of CO.sub.2e emission of each processing step (the carbon hot spot can be identified) [0234] CO.sub.2e certification from SGS [0235] Track sustainability performance over time

[0236] The expert insights system can be realized as a central platform intended for connected products and services, optimizing plant's efficiency, and reducing maintenance times, energy consumption and wastage. Turning machinery into connected devices allows the free flow of data from sensors, machines, and control units to a single secure storage location. Connecting plants to the central platform forming a gateway and harnessing the benefits of digitalization. The central platform allows to bring transparency into processes and machine datae.g. equipped with individual dashboards displaying most important KPIs. Such a transparency enables to provide or initiate or electronically signal concrete actions to increase performance and optimize processes. Thus, connected devices, operational metrics and analytics help optimize efficiency at a plant. Choose from a wide range of digital services to improve productivity, enhance product quality, and reduce wastage and energy consumption. With access via a tablet or smartphone, as embodiment variant, this allows even to control plant on the move.

REFERENCE LIST

[0237] 1 extruder system [0238] 2 decentralized extruder networked system [0239] 101 feeder [0240] 102 extruder [0241] 103 shaping opening [0242] 104 collection mean [0243] 105 operational unit [0244] 111 sensor [0245] 120 Screw [0246] 121 Barrel [0247] 122 conditioning unit [0248] 123 motor [0249] 200 input parameter [0250] 201 operational setting parameter [0251] 202 material characteristics parameter [0252] 203 environmental measuring parameter [0253] 204 target material characteristics parameters [0254] 206 predefined target parameter values [0255] 207 sensory parameter [0256] 208 operational setting parameters [0257] 209 environmental measuring parameter [0258] 211 initial operational setting parameters [0259] 212 digital recipes [0260] 213 Output material [0261] 300 extrudate [0262] 301 input material [0263] 302 Output material [0264] 303 Ingredient [0265] 401 operational setting [0266] 500 extruder control [0267] 501 programable logic [0268] 502 digital controller [0269] 503 repository storage unit [0270] 504 digital database [0271] 505 data transmission network [0272] 506 network interface [0273] 507 data records [0274] 511 operational data storage [0275] 512 nonlinear processing units [0276] 512 central repository [0277] 513 central digital database [0278] 514 central digital controller [0279] 515 operational units [0280] 515 central data records [0281] 601 input layer [0282] 602 hidden layer [0283] 603 output layer [0284] 610 machine learning unit [0285] 611 deep Learning (DL) structure [0286] 612 convolutional layer [0287] 613 pooling layer