Module for sequential data acquisition with integrated AI evaluation

Abstract

The invention relates to a module for sequential data acquisition with integrated AI evaluation, comprising: a measurement unit designed to receive a measurement command containing at least the following measurement parameters: a measurement type, a desired number of measurements, and further designed to sequentially perform the number of measurements according to a particular measurement type of the measurements and to acquire a measurement sequence, wherein the AI microcontroller is designed, starting from a first measurement sequence, to cyclically retrieve the first measurement sequence from the measurement unit and to store the first measurement sequence in an array, and the AI microcontroller is further designed to apply an AI algorithm to measurement values of the first measurement sequence and to output a result via the backplane bus by means of the operating system module, the AI microprocessor is further designed to evaluate a further measurement sequence from the array using an AI algorithm after the evaluation of the first measurement sequence.

Description

Technical Field

[0001] The invention relates to a module for sequential data acquisition with integrated AI evaluation, the acquisition module comprising a measuring unit designed to acquire measured values ​​from a test object via a sensor, an AI microcontroller designed to acquire and process AI algorithms, a data memory, an operating system module, and the module is also designed for a modular structure in an automation system to transmit data to a superordinate unit via a backplane bus. Background Art

[0002] Because AI technology (artificial intelligence, also known as mechanical intelligence) is becoming more and more reliable and is increasingly used in industrial systems, there is a demand for industrial AI equipment. Conventional automation devices such as programmable logic controllers (PLC, also known as SPS in German) do not provide the computing power required for AI technology. Cloud-based AI solutions are available, but they cannot be easily implemented for workshop solutions, that is, the results of cloud-based AI solutions cannot be easily provided in the workshop.

[0003] The trend towards automation systems during the development of digitalization is accompanied by the growth of ever greater volumes of data. One reason for this growth is the integration of processes, which are supposed to monitor and guarantee the quality of the products in the process. Impedance spectroscopy is a measurement method that enables the description of physical phenomena that otherwise cannot be represented by simple, relevant mathematical models. However, interpreting the large amounts of data generated is not trivial, for which empirically based evaluations using artificial intelligence offer a solution. The current trend is to perform AI-based operations in the cloud. For this, the data must be transmitted from the source via the backplane bus, the CPU (central processing unit), and the higher-level control system to the cloud. The data is evaluated in the cloud and sent back via the same path. The workload of the backplane bus generated here is enormous, because such a volume of data is not designed for it. Relocating the analysis to the local peripheral device closest to the data source reduces the communication volume to a minimum and also guarantees data sovereignty for the user. Summary of the invention

[0004] The object of the present invention is therefore to provide a module, in particular for automation technology, which allows the user to evaluate the measurement data acquired for his process-specific requirements on site with the aid of an AI network or a neural network, without using a cloud service.

[0005] This object is achieved in that the measuring unit is designed to receive a measurement command containing at least the following measurement parameters: a measurement type, a desired number of measurements, and the measuring unit is also designed to perform the number of measurements in sequence according to a specific measurement type and to obtain a measurement sequence, wherein the AI ​​microcontroller is designed to start from a first measurement sequence, cyclically retrieve the first measurement sequence from the measuring unit and store the first measurement sequence in an array, and furthermore, the AI ​​microcontroller is also designed to apply an AI algorithm to the measured values ​​of the first measurement sequence and output the results via a backplane bus with the aid of an operating system module, and the AI ​​microcontroller is also designed to evaluate another measurement sequence from the array using the AI ​​algorithm based on the evaluation of the first measurement sequence.

[0006] According to the invention, freely programmable measuring units and freely programmable AI microcontrollers, i.e. AI evaluations, are now integrated into modules, in particular peripheral modules for modular construction, such as local peripheral modules. Users can now parameterize their measuring units, for example for structure-borne sound measurements or vibration measurements or impedance spectroscopy, and make the acquired results available directly for evaluation with the aid of AI algorithms.

[0007] All data can now be preprocessed and only the results, such as classification results from a measurement sequence, can be transmitted in order to reduce the data load. In particular, the use of modules can ensure the real-time capability of the master controller in automation technology by reducing the load on the backplane bus, thus reducing the load on the master controller and thus shortening the overall delay time.

[0008] Another advantage of such modules is the adaptability of the modules to the most diverse application cases and thus the simplification for the commissioning engineer. Standardized hardware is now available which can solve production engineering tasks with correspondingly parameterized and adapted neural networks.

[0009] According to the invention, a neural network, i.e. an artificial neural network, is a neural structure mapped with learned knowledge. From the field of artificial intelligence there are methods of "knowledge discovery in databases", which are also often referred to as data mining. From this data mining comes the general requirement of discovering unknown connections from mostly very large data sets. Here, the operation of the algorithms used is different, i.e. there are no prescribed professional result expectations for analyzing the data. For supervised learning methods, such as classification, specific results are expected. Deviations can be trained with the model so that it reacts more robustly to severe fluctuations. How to react to fluctuations depends on the algorithm, the parameters therein and the data with which it is trained. In particular, materials in automated production are subject to production-related or batch-related fluctuations.

[0010] In this module, the measuring unit and the AI ​​microcontroller are advantageously designed to be parameterized for the corresponding measurement type by means of an engineering station and are thus parameterized for structure-borne sound measurements, vibration measurements or impedance measurements.

[0011] In particular, in parameterization for impedance measurement, the measuring unit is designed as an impedance measuring unit and acquires a measurement sequence or an impedance spectrum dependent on the frequency and / or amplitude of the test signal from a test object via a sensor using an electrical test signal, wherein the impedance measuring unit is designed to start with a start frequency by means of a measurement command and to perform a last measurement at a stop frequency, thereby acquiring the desired number of measurements.

[0012] This has the advantage of a uniform distribution of the frequencies over the entire measuring frequency band. In order to expand the measuring range in ratio measurements, four calibration resistors can optionally be switched for the impedance measuring unit via a multiplexer. After the measurement at a specific frequency is completed, the next frequency is calculated according to the following formula. This formula is then also parameterized in the measuring unit or in the microcontroller.

[0013]

[0014] in:

[0015] K - total number of measurements

[0016] k - the index of the current measurement

[0017] Also implemented in the measuring unit or microcontroller is corresponding control of the multiplexer, wherein the impedance measuring unit or AI microcontroller is designed to start the measurement with the value of the calibration resistor and evaluate the measured data, where a comparison is made as to whether the measured impedance is less than 80% of the calibration resistor, and if so, the multiplexer switches on the next smallest calibration resistor.

[0018] For this purpose, the same signal can now be applied to the unknown DUT and the unknown reference resistor. In the resistance measurement using the reference resistor, the unknown resistance, i.e. the impedance, is determined using a voltmeter by means of the voltage drop between the DUT and the reference resistor. Because the reference resistor is known, the unknown resistance can be determined by multiplying the ratio of the voltage by the reference resistor. The resulting ratio equation can be expressed in C code and used to calculate the unknown impedance of the DUT. Due to the measurement process, if the difference between the impedance of the calibration resistor and the DUT is too large, the accuracy of the measurement will be greatly affected. Therefore, each frequency of the measurement signal is evaluated and the measured data is evaluated. What is compared here is whether the measured impedance is less than 80% of the reference resistor. If so, the microcontroller switches to the next smaller reference resistor with the help of a multiplexer and repeats the measurement.

[0019] Especially in impedance measurements, surprising results in terms of accuracy occur, because if the AI ​​microcontroller is designed to apply the AI ​​algorithm to the imaginary part of the measured values ​​of a measurement series, accurate results or classifications can be achieved by the AI ​​algorithm. The measuring unit can be parameterized to measure the physical variables: amplitude, phase, real part, imaginary part and loss factor. Thus, each measurement series can contain up to five parameters, or just one or any combination.

[0020] In order to make the evaluation in the module more robust to unknown data, the AI ​​microcontroller is designed to evaluate the AI ​​algorithm using a residual neural network. Preferably, a residual network with a total of 12 convolutional layers is used, i.e., typical residual neural networks have a depth of 18, 34, 50, 101 and 152 convolutional layers, but it has been shown that 12 layers are sufficient for the given application, because the module converges within a few training cycles.

[0021] In order to advantageously use the module in an automation system, the operating system module is designed so that the AI ​​microcontroller and the measuring unit can be parameterized or programmed via the backplane bus. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings illustrate embodiments of the present invention, wherein

[0023] Figure 1 is a module designed for sequential data acquisition with integrated AI evaluation,

[0024] Figure 2 is an array for measuring cells,

[0025] Figure 3 It is the principle structure of AI microcontroller, and

[0026] Figure 4 There are possible flows for the initialization and measurement flow of the AI ​​processor and the measurement unit. DETAILED DESCRIPTION

[0027] according to Figure 1A module 1 for sequential data acquisition with integrated AI evaluation is shown. The module 1 is divided into a base board BB and a measuring module MM. The base module BM has a power supply unit SV, which supplies a 24 volt voltage P24V and a ground connection M24 from the outside. In addition, the power supply unit SV has a transformer, which emits a constant 5 volt voltage. For general control and coordination purposes, the base board BB and the measuring module MM are involved, and the base board BB has an operating system module 7. The operating system module 7 is connected to the backplane bus interface 8 on the one hand, and is connected to the isolation module 13 via internal data communication lines, such as GPIO1 to GPIO9 (General Purpose Input Output) for electrical isolation of signals on the other hand. The isolation module 13 is designed as a magnetic coupler in this case. Accordingly, the backplane bus interface 8 of the operating system module 7 is correspondingly connected to the backplane bus RWB. The backplane bus RWB is used to obtain other modules, input / output components, technical components, etc. in a modularly constructed automation system. In addition, the backplane bus RWB connects the mentioned modules to the upper-level programmable logic controller.

[0028] The measuring module MM integrated into the module 1 has a power supply connection P5V on the one hand and a data connection SPI on the other hand to the basic module. The first voltage regulator 4 on the measuring module MM provides a constant voltage of 1.8 volts for the subsequent modules. The second voltage regulator 5 provides a constant voltage of 3.3 volts for the subsequent modules.

[0029] The core of the measurement module MM is a freely programmable AI controller 3 and a likewise freely programmable measuring unit 2. The AI ​​microcontroller 3 and the measuring unit 2 are likewise connected to each other via an internal data line, namely an SPI (Serial Peripheral Interface). The measuring unit 2 is designed to acquire measured values ​​from a test object 40 via a sensor 6. The AI ​​microcontroller 3 is designed to acquire and store AI algorithms. The data memory 9 can be used to cache measurement sequences. The measuring unit 2 is designed to receive a measurement command V containing at least the following measurement parameters. Measurement types KM, SM, IM, wherein KM should correspond to structure-borne sound measurement, SM should correspond to vibration measurement, and IM to impedance measurement according to the impedance spectroscopy principle. This means that the measuring unit 2 can be freely programmed for a specific application via a predefined measurement command V. In addition, the measuring unit 2 is informed of the number K of measurements to be performed in sequence. In this case, the AI ​​microcontroller 3 is designed to start from a first measurement sequence M1, cyclically retrieve the first measurement sequence from the measuring unit 2 and store the first measurement sequence in an array A. The dashed line between the AI ​​microprocessor 3 and the measuring unit 2 indicates that, when programming, for example, the type of KM corresponding to structure-borne sound measurement, the AI ​​microprocessor 3 and the measuring unit 2 are programmed or parameterized together with one another in order to interact accordingly with the neural network involved in the subsequent measurement and evaluation process.

[0030] The AI ​​microcontroller 3 is designed to apply the AI ​​algorithm to the measured values ​​of the first measurement sequence M1, of course also the measured values ​​of the subsequent measurement sequences, and to output the results via the backplane bus RWB with the aid of the operating system module 7. It is particularly advantageous in this design that the AI ​​evaluation is performed directly after the acquisition of the measurement sequence into the AI ​​microcontroller 3. This enables fast, memory-saving sequential processing on site. In addition, the AI ​​microcontroller 3 is designed to evaluate the further measurement sequences M2, ..., M10 from the array A using the AI ​​algorithm after evaluating the first measurement sequence M1.

[0031] This has the advantage that the AI ​​microcontroller 3 is designed for parameterization for the respective measurement type KM, SM, IM by means of an energy station and for parameterization for structure-borne sound measurement, vibration measurement or impedance measurement.

[0032] If the AI ​​microcontroller 3 and the measuring unit 2 are parameterized for impedance measurement, the measuring unit 2 is designed to acquire a measurement sequence or an impedance spectrum IMS dependent on the frequency and / or amplitude of the test signal PS from the test object 40 via the sensor 6 using an electrical test signal PS also generated by the measuring unit. The measuring unit 2 or the impedance measuring unit parameterized in this way is designed to start with a start frequency f and to stop with a stop frequency f by means of a measurement command V and to thereby acquire the desired number K of measurements.

[0033] The particularity in the parameterization of the measuring unit 2 is that the impedance measuring unit and / or also the AI ​​microcontroller 3 is designed to start the measurement and evaluate the measured data with the value of the calibration resistor 12 or Rcalibrate and to compare whether the measured impedance is less than 80% of the calibration resistor, and if so, the multiplexer 11 switches to the next smallest calibration resistor.

[0034] In order to realize corresponding frequency stepping starting from the start frequency f and ending at the stop frequency f, there is also a frequency calculation formula 81 in the impedance measurement unit 2, which predefines the logarithmic frequency stepping, and linear predetermination can also be realized.

[0035]

[0036] Furthermore, the measuring module MM has a memory card 10 for additional programs or for acquiring measurement files. Optionally, an array A for the measurement sequences M2 , . . . , M10 can also be stored in the data memory 9 .

[0037] To explain more clearly, Figure 2 Array A is shown again, in which the first measurement sequence M1, the second measurement sequence M2 up to the tenth measurement sequence M10 are stored in succession. For the final evaluation by the AI ​​microcontroller 3, array A can now be processed in sequence one after the other and can be evaluated by the AI ​​algorithm one measurement sequence after the other. In order to always be able to transfer a new measurement sequence from the measuring unit 2 to the AI ​​microcontroller 3 and thus into the array A, the measuring unit 2 informs the AI ​​microcontroller 3 via the interrupt IRQ, and the AI ​​microcontroller 3 then retrieves the measurement sequence to be evaluated from the measuring unit.

[0038] Figure 3 The general structure of an AI microcontroller is shown. An assembled chip 30 is shown as an example. An on-chip system memory 32 having 512KB flash memory and 128KB of SRAM (static random access memory) is provided on the chip 30. Similarly, a communication interface 34 such as Quad SPI (Quad Serial Peripheral Interface), I2S (Inter-Integrated Circuit Sound Bus) or a camera interface PCIF is provided. In order to reduce the energy consumption of the chip 31, it has an ARM Cortex M4F-M4 with a CPU for system control and a Risk-V processor for customer-specific applications. In order to speed up the completion of the AI ​​algorithm process, the chip 30 also has an accelerator 33. The accelerator 33 is also called a CNN (Convolutional Neural Network) accelerator and is capable of running multiple processors in parallel. It is also capable of acquiring 32-64 for AI evaluation. 3 multiple layers; it is able to acquire weights; it has a maximum number of input and output variables, etc.

[0039] The main component of the AI ​​microprocessor 3 is therefore the CNN accelerator. The accelerator has 64 processors, divided into four quadrants, each with 16 processors. Each processor is connected to its own weight memory instance, and four processors share one data storage instance. In general, each quadrant has its own data memory, weight memory and bias memory. The data memory for all quadrants can be as shown in Figure 3 All input data that should be processed in the neural network must be stored in the corresponding data memory instance of the corresponding processor.

[0040] Figure 4 A program operation plan for possible parameterization and measurement execution of the AI ​​microprocessor 3 and the measuring unit 2 is shown. Starting with start 40, the AI ​​microprocessor 3 is first programmed in an initialization step 41. The measuring unit 2 is initialized in another initialization step 42. In a third initialization step 43, the maximum resistance on the multiplexer 11 is first set in the example for impedance measurement. The AI ​​microprocessor 3 is then started in this way and the CNN accelerator is turned on in a fourth step 44. The CNN accelerator is initialized in a fifth step 45. The weights are then loaded into the weight memory of the CNN accelerator in a sixth step 46. In a seventh step 47, the bias value is also loaded into the bias memory. The CNN accelerator is now configured for a given model in an eighth step 48. The AI ​​microprocessor 3 is now ready to process input data and loads the input data in a ninth step 49.

[0041] Now it must be decided via a no-answer test 50 whether to start with a sample data set or to start with loading a real measurement sequence. In the case of a sample data set, the sample data set is loaded by executing the loading of the sample data set 51 and the evaluation of the sample data set 52 is started. A check 53 is performed based on the sample data set to determine whether the processing and evaluation were carried out correctly; if so, the inference output 54 is loaded and the output probabilities are calculated in a further step 55. Then, the CNN accelerator is closed in a step 56 and the classification result 57 is displayed.

[0042] For the alternative of loading the array of the actual measurement sequence, the array is loaded in a loading step 60, then the reasoning 61 is started and the reasoning output is compared with the given result in a check 62. If this is successful, the reasoning output 63 is loaded and the output probability 64 is also calculated. The CNN accelerator 65 is then closed and the classification result 66 is displayed. The measuring unit 2 can now be initialized and the measurement results are written to the array A again in step 68. Scaling of the array A from -128 to +127 is performed. Subsequently, the categories can be classified again in the AI ​​microcontroller 3 in step 70. It is now important that the results are written to the cache in step 71 and that the operating system module 7 can therefore read out in a read step 72 and provide these results on the backplane bus RWB. The measurement process is completed with an end 73.

Claims

1. A module (1) designed for sequential data acquisition with integrated AI evaluation, comprising A measuring unit (2) is designed to acquire measured values ​​from a test object (40) via a sensor (6), AI microcontrollers (3), designed to acquire and process AI algorithms, Data storage (9), Operating system modules (7), The modules are also designed for modular construction in automation systems to transfer data to higher-level units via the backplane bus (RWB). It is characterized in that The measuring unit (2) is designed to receive a measuring command (V) containing at least the following measuring parameters: Measurement type (KM, SM, IM), The desired number of measurements (K), The measuring unit is also designed to perform the number (K) of measurements in sequence according to a specific measurement type (KM, SM, IM) and to acquire a measurement sequence (Mi), wherein the AI ​​microcontroller (3) is designed to start from a first measurement sequence (M1), cyclically retrieve the first measurement sequence from the measuring unit (2) and store the first measurement sequence in an array (A), the AI ​​microcontroller (3) is also designed to apply the AI ​​algorithm to the measured values ​​of the first measurement sequence (M1) and output the results via the backplane bus (RWB) with the aid of the operating system module (7), and the AI ​​microcontroller is also designed to evaluate further measurement sequences (M2, . . . , M10) from the array (A) using the AI ​​algorithm after evaluating the first measurement sequence (M1).

2. The module (1) according to claim 1, wherein: The measuring unit (2) and the AI ​​microcontroller (3) are designed to carry out parameterization for the respective measurement type (KM, SM, IM) by means of an energy station and for structure-borne sound measurement, vibration measurement or impedance measurement.

3. The module (1) according to claim 1 or 2, wherein: In parameterization for impedance measurement, the measuring unit (2) is designed as an impedance measuring unit and is therefore able to acquire, by means of an electrical test signal (PS) via the sensor (6), a measurement sequence or an impedance spectrum (ImS) from the test object (40) which is dependent on the frequency and / or amplitude of the test signal (PS), wherein the impedance measuring unit is designed to start with a start frequency (fstart) and end with a stop frequency (fstop) by means of the measurement command (V) and thereby acquire the desired number (K) of the measurements.

4. The module (1) according to claim 3, comprising a multiplexer (11), wherein: The impedance measurement unit or the AI ​​microcontroller (3) is designed to start measuring with the value of the calibration resistor and evaluate the measured data and compare whether the measured impedance is less than 80% of the calibration resistor. If so, the multiplexer switches to the next smallest calibration resistor.

5. The module (1) according to claim 3 or 4, wherein: The AI ​​microcontroller (3) is designed to apply the AI ​​algorithm to the imaginary part of the measured values ​​of the measurement sequence.

6. The module (1) according to any one of claims 1 to 5, wherein: The AI ​​microcontroller (3) is designed to evaluate the AI ​​algorithm using a residual neural network.

7. The module (1) according to any one of claims 1 to 6, wherein: The operating system module (7) is designed so that the AI ​​microcontroller (3) and the measuring unit (2) can be parameterized or programmed via the backplane bus (RWB).

Images