Monitoring method, computer program product, monitoring unit, gas analysis device and use of artificial intelligence

Abstract

The invention relates to a monitoring method (200) for a system (10) having a plurality of devices (12) which are designed to provide associated measured values (15) and / or control commands (17) in each case. The monitoring method (100, 200) evaluates the measured values (15) or the control commands (17) by means of another processing device (40) or a neural network (80) on the basis of the use thereof. The invention further relates to a computer program product (70), by means of which the monitoring method (100, 200) can be carried out, and to a correspondingly equipped monitoring unit (75). The invention further relates to a gas analysis device (11) equipped with such a monitoring unit (75). The invention further relates to the use of an artificial intelligence (99) for monitoring a gas analysis device (11).

Description

Technical Field

[0001] The invention relates to a monitoring method for a monitoring system and a computer program product designed therefor. The invention also relates to a monitoring unit with a corresponding computer program product and a gas analysis device with such a monitoring unit. The invention also relates to the use of artificial intelligence for monitoring a gas analysis device. Background Art

[0002] Document KR 2021147318 A discloses a multi-sensor artificial intelligence for fault diagnosis of mechanical equipment, wherein sensor data is collected and processed by an autoencoder.

[0003] Application US2020 / 0310370 A1 discloses a control system with an autoencoder, wherein sensor data is processed into derived sensor data. The autoencoder receives the derived sensor data and makes predictions from them. Based on the predictions, a control intervention is triggered.

[0004] Complex systems with an increasing number of individual devices are increasingly used in different applications of automation technology. Likewise, flexible operation is intended to be achieved in order to process different input materials. Due to the increasing complexity of the operation and the systems, the monitoring of such systems is becoming more complex. Therefore, there is a need for a method that enables the monitoring of complex systems, provides reliable automatic fault acquisition and can quickly adapt to changing operating requirements. This applies in particular to gas analysis devices, which are becoming increasingly complex and sensitive due to the increasing requirements on measurement accuracy. Summary of the invention

[0005] This object is achieved by a kind of innovative monitoring method, and this monitoring method is designed to monitor the operation of system.This system comprises a plurality of equipments, and these equipments are respectively designed to provide measured value and / or control command.Equipment can be sensor, control unit, adjustment unit or its combination, for example.Measured value can be recorded physical quantity or state information about corresponding equipment.Monitoring method comprises a first step, in this step, an input signal array is provided, and this array comprises a plurality of units.The unit of input signal array is suitable for receiving predetermined measured value or control command, and transmits to another processing unit.The input signal array can be constructed into one-dimensional, two-dimensional or higher dimensional array, i.e. data field.In the first step, also provide monitoring data array, it corresponds with input signal array in dimension and size (i.e. number of units).The unit of input signal array corresponds to the unit of monitoring data array respectively.

[0006] Likewise, the monitoring method has a second step in which the system is placed in an active operating state. In the active operating state, measured values ​​or control commands are essentially continuously generated and prepared for further processing of the monitoring method. The acquired measured values ​​and / or control commands are transferred to the input signal array and input as content into the corresponding cells. These contents are processed by a further processing unit into the content of the corresponding cells in the monitoring data array.

[0007] In the third step, a system deviation parameter is determined based on the input signal array in the second step. The system deviation parameter consists of a plurality of cells of the monitoring data array. The system deviation parameter can, for example, consist of the sum of the contents of the cells of the monitoring data array or the sum of the squares of the contents of the cells of the monitoring data array. The cell contents of the monitoring data array can be standardized to a reference value individually or together. These contents are compatible with the content type of the corresponding cells in the input signal array. It is also possible to weight the individual cells of the monitoring data array. The weighting can be performed by an algorithm before or during the execution of the monitoring method. If in the third step the system deviation parameter exceeds an adjustable system threshold, the presence of an abnormal state is obtained.

[0008] In addition, the monitoring method also includes a fourth step, which can be performed when an abnormal operating state of the system is obtained in the third step. In this step, the content of the monitoring data array unit combined in the system deviation parameter is further processed separately. According to the content of the monitoring data array unit and the content of the corresponding unit of the input signal array, the unit deviation parameter is determined in pairs. For example, this can be achieved by calculating the difference. Next, in the fourth step, when the value of the associated unit deviation parameter exceeds an adjustable unit threshold, the unit of the input signal array or the monitoring data array will be identified as a defective unit. The unit threshold can be set by a user, a table, an algorithm or artificial intelligence. The defective unit can be a unit of the input signal array or a unit of the monitoring data array. Therefore, the device to which it belongs can be identified as defective. In particular, a certain component of the device can also be identified as defective. In addition, in the monitoring method of the present invention, when a defective unit is obtained, a warning will be issued. The warning can be a visual or auditory warning to the user, and / or as an electronic warning signal, for example, sent to the control unit of the device. Through the electronic control signal, the control unit can initiate countermeasures in the operation of the system.

[0009] The monitoring method of the present invention is basically composed of two stages. First, the presence or absence of an abnormal operating state is identified by the second and third steps. Only when an abnormal operating state is obtained, the defective unit is identified by recording and analyzing the individual unit deviation parameters, thereby obtaining detailed information on the current abnormal operating state. As long as a normal operating state exists, the third step only compares between the system deviation parameter and the system threshold. Therefore, under normal operating conditions, the number of computationally intensive comparison operations that need to be performed is minimized. The system deviation parameter can, for example, be designed as the sum of the unit deviation parameters, which can be determined by reducing the computing cost. Only when an abnormal operating state is obtained, more comparison operations are performed using the unit deviation parameters and the associated unit thresholds. Therefore, the monitoring method of the present invention can be performed by reducing computing power. Therefore, the monitoring method according to the present invention, in particular the second, third and fourth steps, provides real-time capabilities for a wide range of systems. The concept of real-time capabilities should be understood here according to the specific meaning of the corresponding device. The present invention is based in part on the following recognition: in the third step, the system deviation parameter can be reliably used as a decision criterion for executing the fourth step. According to the monitoring method of the present invention, reliability is improved while reducing computing costs as a whole.

[0010] In one embodiment of the claimed monitoring method, a fifth step is included, wherein the defective unit or the determined defective unit group determined in the fourth step is further utilized. In the fifth step, the defective unit or the defective unit group is recorded and evaluated by pattern matching. By pattern matching, the cause of the defect is determined in the fifth step. For this purpose, a database can be used, in which a large number of comparison patterns of different damage states, i.e. abnormal operating states of the system, are stored. Alternatively or additionally, the pattern matching can also be performed by artificial intelligence, such as a machine learning algorithm trained before and / or during the operation of the system. In the pattern matching, the extent to which one or more defective units exceed the threshold value of the relevant unit, i.e. the value of the excess, can also be considered. As the cause of the defect, a defective device of the system, a loss of function of such a device or an external influence on the system can be identified. Since the claimed monitoring method can be performed with reduced computing power or more quickly with increased computing power, this artificial intelligence can be further developed, in particular further trained, during the operation of the system. The claimed monitoring method therefore provides improved acquisition of abnormal operating states of the system.

[0011] Furthermore, in the required monitoring method, the first, second, third and / or fourth steps can be performed by an autoencoder. The system deviation parameter determined by the autoencoder can be a so-called total reconstruction error. The autoencoder has a variety of configurations with high reliability and robustness. In addition, the autoencoder is powerful and allows a large amount of input data, i.e., a large number of input signal arrays, to be processed quickly. Therefore, the required monitoring method can be simply implemented and applicable to different systems. In addition, the autoencoder allows multiple steps of the required monitoring method to be performed. In addition, the autoencoder is also capable of so-called unsupervised learning, also known as unsupervised learning. The data set used to train the autoencoder is managed by simplified sorting and reduced data labels. In addition, the autoencoder utilizes the relationship between the existing current data under predetermined operating conditions. Therefore, the use of a large database is redundant during most of the operating time.

[0012] In addition, in the claimed monitoring method, at least one adjustable cell threshold can be set as a fixed threshold or a moving threshold. Fixed thresholds provide simple configurability of the monitoring method. For cell thresholds normalized to reference values, the claimed monitoring method is configured in a simple manner by at least one fixed threshold, for example at initialization. On the other hand, moving thresholds allow for consideration of degradation effects on the system, for example, thereby affecting the corresponding cells of the input signal array. For example, the cell threshold can rise or fall along a function according to the operating time of the system operation. This can avoid inaccurate warnings and / or reliably obtain early degradation of the system. The diagnostic accuracy of the monitoring method used is thus further improved. For example, the threshold can be an adjustable multiple of the standard deviation of the corresponding data and / or an adjustable multiple of the average total reconstruction error. Therefore, a fixed or sliding threshold can be provided. Whether the threshold is sliding or fixed depends on whether a base value, such as a standard deviation or an average total reconstruction error, is carried during the claimed monitoring method.

[0013] In another design of the claimed monitoring method, the second step can include interpolating the measured values ​​and / or control commands and passing them to the input signal array. Therefore, the input signal array provides synchronized measurement points so that each input of the input signal array describes the operating state of the system within a specific discrete time interval. In this way, a higher time resolution can be achieved in the claimed monitoring method. Therefore, in the transient operating state of the system, ambiguities or contradictions in the input signal array, as well as incorrect warnings caused thereby, are minimized. The interpolation can be designed as a so-called upsampling. Therefore, the claimed monitoring method is suitable for time-critical applications, in which a fast response to the current operating state is required. The claimed monitoring method therefore has a higher reliability and diagnostic accuracy. Alternatively, so-called downsampling can also be performed in the second step. Therefore, only those measured values ​​and / or control commands that are already within the discrete time interval are passed to the input signal array. As a result, the computing power required for the claimed monitoring method is reduced.

[0014] Furthermore, the input signal array and / or the monitoring data array can contain cells, in which measured values ​​and / or control commands satisfying an adjustable relevance criterion are stored during the claimed monitoring method. The relevance criterion can be an indication of how many known defects are expected to exceed a cell threshold in the respective cell. A cell which numerically exceeds the cell threshold with a minimum number of known defects, in particular with no known defects, can be removed from the input signal array. Thus, the claimed monitoring method can be automatically further optimized with increasing operating time and thus accelerated. This allows the system to be easily modified without affecting the performance of the claimed monitoring method. The relevance criterion can in particular be used for so-called pruning, for example pruning of the input signal array.

[0015] Another monitoring method according to the present invention also achieves the basic purpose. The monitoring method is used to monitor a system having a plurality of devices that interact to achieve the expected operation of the system. The devices can be sensors, control units, regulating units or combinations thereof. The devices are designed to form nodes in a neural network that maps the interaction of the devices in the system. The neural network includes zero correlation and / or operational correlation. The devices can be physically independent devices or different functions in a physically single device. The monitoring method includes a first step, in which a trained neural network is provided, the network having a plurality of edges, each edge corresponding to a zero correlation or an operational correlation between two nodes in the neural network. In this case, the nodes are formed by the devices of the system. Zero correlation refers to a correlation whose correlation value is below an adjustable correlation threshold. The so-called zero correlation describes a weak, negligible or non-existent relationship between two nodes. The term zero correlation should be understood according to the unpublished German patent application with the official file number 102022200 694.1. The content of the document 10 2022 200 694.1 is incorporated into the present application by reference. In the provided trained neural network, zero correlation has a correlation value that is below an adjustable correlation threshold. When training the neural network, the corresponding correlation value is determined. Thus, zero correlation describes that there is no expected significant correlation between the device pairs in a specific operating state. Operating correlation refers to the existence of a sufficiently strong correlation between two nodes. Therefore, operating correlation is the opposite or counterpart of zero correlation. The term operating correlation should be understood in accordance with document 10 2022 200 694.1 in correspondence with the term zero correlation.

[0016] Furthermore, the monitoring method comprises a second step in which the system is in an active operating state and records measured values ​​and / or control commands occurring in the active operating state. The measured values ​​can be recorded physical quantities or status information about individual devices. The recorded measured values ​​and / or control commands correspond to at least one node and / or edge of the neural network, respectively. The recorded measured values ​​and / or control commands are suitable as inputs to the neural network.

[0017] In a third step, a network deviation parameter is determined, which combines the correlation values ​​of multiple edges of the neural network. To this end, the correlation values ​​of multiple edges are determined based on the measured values ​​and / or control commands recorded in the second step, for example, summarized in a sum by magnitude. The correlation values ​​of the edges can be recorded, obtained from zero correlation and / or operating correlation, to determine the network deviation parameter. Therefore, the network deviation parameter represents a compressed quantity, reflecting the overall state of the system. This is followed by a fourth step, when the network deviation parameter exceeds an adjustable network threshold in value, an abnormal operating state of the system is identified. The network threshold can be preset by the user or by an algorithm and represents a measure reflecting the degree to which the device deviates from the expected target state. Therefore, the network threshold is exceeded or exceeded due to an increase in the correlation values ​​of the edges, which should have zero correlation, so that they can no longer be correctly classified as zero correlation individually or in combination. Therefore, the monitoring method of the present invention can identify false correlations that should or actually have zero correlations in a specified operating state. Alternatively, the operating correlations may also deviate, and these deviations combined may lead to exceeding the network threshold. The monitoring method of the invention is based on the surprising discovery that abnormal operating states can be identified in a simple and reliable manner by zero correlations, which otherwise have no particular significance for the operation. A detailed understanding of the zero correlations is superfluous, since they do not rely on any underlying physical connection. In addition, edges with zero correlations can be generated essentially automatically in number. Likewise, deviations from operating correlations can be easily detected and abnormal operating states can be identified. If an abnormal operating state is detected, a warning will be issued. The warning can be a visual or audible warning to the user and / or an electronic warning signal, for example sent to a control unit of the system. By means of the electronic control signal, the control unit can initiate countermeasures in the operation of the system.

[0018] The network deviation parameter can be provided at a low computational cost and checked by a minimum of comparison operations whether it exceeds the value of the network threshold. Therefore, the monitoring method of the present invention can be executed at a low computational cost or at a higher speed. Through the first, second, third and fourth steps, the real-time capability of the monitoring method of the present invention is improved. Here, the concept of real-time capability should be understood according to the specific application, that is, the system monitored by the monitoring method. At the same time, the monitoring method allows reliable diagnosis of existing abnormal operating states.

[0019] According to a design of the monitoring method claimed, the second, third and / or fourth steps can be performed for zero correlation and operation correlation, respectively. In particular, the second, third and / or fourth steps can be performed with an adjustable first frequency for zero correlation. In addition, the second, third and / or fourth steps can be performed with an adjustable second frequency for operation correlation. The first or second frequency can be set by a user or an algorithm. In addition, the first frequency can be lower than the second frequency. Therefore, the monitoring method claimed can adapt to different applications, wherein abnormal operating states are expected to be more easily identified by operation correlation or zero correlation. In general, the monitoring method claimed has a wide range of applications. A further alternative is that the second, third and / or fourth steps can be performed in the first process for operation correlation. If an abnormal operating state is obtained in the first process, the second, third and / or fourth steps can be performed for zero correlation. Determine a separate network deviation parameter in the first and second processes and compare it with a corresponding adjustable network threshold. The monitoring method claimed can be implemented with lower computing power.

[0020] In another design of the claimed monitoring method, a fifth step is included, in which an edge of the neural network that corresponds to zero correlation in the normal operating state of the system is identified as a defective edge. When the correlation value of the edge absolutely exceeds the corresponding correlation threshold, the edge is identified as a defective edge. The fifth step is performed when an abnormal operating state of the system is identified in the fourth step. The absolute exceeding of the correlation threshold is achieved by comparing the correlation value with the corresponding correlation threshold separately. Therefore, only when at least one correlation threshold is expected to be exceeded, a computationally intensive comparison operation is performed based on the exceeded network threshold, thereby avoiding frequent resultless computationally intensive operations. Therefore, the monitoring method of the present invention provides a method for rational use of computing power. At the same time, the claimed monitoring method provides targeted diagnosis.

[0021] Furthermore, the correlation threshold of at least one edge in the neural network can be set as a fixed threshold or as a moving threshold. Fixed thresholds provide for simple configurability of the monitoring method. For correlation thresholds that are normalized to reference values, the claimed monitoring method can be configured in a simple manner by means of at least one fixed threshold, for example at initialization. On the other hand, moving thresholds allow for taking into account, for example, degradation effects on the device, thereby affecting the corresponding edge of the neural network. For example, the moving threshold can be formed by a variable total reconstruction error. For example, the correlation threshold can rise or fall along a function depending on the operating time of the system. This makes it possible to avoid inaccurate warnings and / or reliably detect premature degradation of the system. The diagnostic accuracy of the claimed monitoring method is thus further improved.

[0022] Furthermore, the claimed monitoring method can include a sixth step, in which pattern matching is performed based on the defect edge identified, for example, in the fifth step. In the sixth step, the defect edge is acquired and evaluated by pattern matching. By pattern matching, the cause of the defect is determined in the sixth step. For this purpose, a database can be used in which a large number of comparison patterns of different damage states, i.e. abnormal operating states of the system, are stored. Alternatively, pattern matching can also be supplemented or replaced by artificial intelligence, such as a machine learning algorithm trained before and / or during the operation of the system. In pattern matching, the extent of one or more defect edges can also be taken into account, i.e. the extent to which these edges numerically exceed a correlation threshold. As the cause of the defect, a failure of the system, a loss of such a system function or an external influence on the system can be identified. Since the claimed monitoring method can be executed at a reduced computing power or executed more quickly at an increased computing power, such artificial intelligence can be further developed as the system is operated, in particular can be further trained. The claimed monitoring method thus provides an improved acquisition of abnormal operating states of the system.

[0023] The described purpose is also achieved by an innovative computer program product. The computer program product is designed to receive measured values ​​and / or control signals in a device, store and process these signals at least temporarily. The computer program product is also designed to issue warnings and / or electronic warning signals to a user. The computer program product is used to monitor a system, including the operation of the device and their interaction. The measured values ​​and / or control commands are processed in a monitoring method, which can be carried out simultaneously with the operation of the system, essentially in parallel, especially in real time. According to the present invention, the computer program product is designed to perform at least one of the monitoring methods of the above-mentioned embodiments. The computer program product can be monolithic, i.e. designed to run on a single hardware platform. Alternatively, the computer program product can be modular, i.e. contain subroutines running on separate hardware platforms, such as a computer cloud, and interact via a communication data connection to perform respective monitoring methods. The computer program product according to the present invention allows the monitoring method of the present invention to be implemented in a simple manner.

[0024] Likewise, the basic purpose is achieved by a monitoring unit for a system according to the invention. The monitoring unit is designed to monitor the operation of a system to which it is at least functionally coupled and to issue a warning. The system to be monitored comprises a plurality of devices which interact to operate the system and are capable of acquiring measured values ​​and / or control commands. The acquisition, including reception, of the measured values ​​and / or control commands is achieved by the monitoring unit. Likewise, the monitoring unit is designed to process the acquired measured values ​​and / or control commands. According to the invention, the monitoring unit is designed to process the measured values ​​and / or control commands by means of a computer program product with which it is equipped. According to the invention, the computer program product is designed according to one of the above-described embodiments.

[0025] The above-mentioned object is achieved by a gas analysis device according to the present invention. The gas analysis device comprises a plurality of devices which interact to adjust and measure a material sample. In particular, the gas analysis device can be designed to at least qualitatively determine the composition of the material sample. The gas analysis device is equipped with a monitoring unit which is connected to the plurality of devices. The monitoring unit is designed to identify abnormal operating conditions of the gas analysis device. According to the present invention, the monitoring unit is designed according to one of the above-mentioned design solutions.

[0026] In addition, the following objects are achieved by the application according to the invention of an artificial intelligence, which is designed to monitor a gas analysis device. The artificial intelligence is suitable for directly or indirectly receiving and processing measured values ​​and / or control commands from the equipment of the gas analysis device. The application of the artificial intelligence at least includes issuing a warning, which can be a visual or audible warning and / or an electronic warning signal issued to the user. In order to monitor the gas analysis device, the artificial intelligence is designed to perform a monitoring method. According to the present invention, the monitoring method can be designed according to one of the above-mentioned embodiments. The application of the artificial intelligence can especially include running a computer program product designed accordingly as described above. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The invention will be described in more detail below by means of various embodiments in the accompanying drawings. These figures should be understood to be complementary to each other, since the same reference numerals in different drawings have the same technical meaning. The features of the various embodiments can also be combined with each other. In addition, the embodiments shown in the accompanying drawings can be combined with the features outlined above. The specific drawings show:

[0028] Figure 1 The operating flow of the first embodiment of the claimed monitoring method is schematically shown;

[0029] Figure 2 The operating flow of the second embodiment of the claimed monitoring method is schematically shown. DETAILED DESCRIPTION

[0030] A first embodiment of the monitoring method 100 claimed is Figure 1 Schematically shown in FIG. The monitoring method 100 is used for monitoring a system 10, which executes a system process 20. The system 10 is designed as a gas analysis device 11, and the system process 20 executed thereon comprises determining the composition 25 of a material sample 22. For this purpose, the system 10 is equipped with a plurality of devices 12, which are designed as sensors 14 or control units 14. The control unit 16 is designed to influence the system process 20 by executing a control program 18. The sensor 14 acquires physical quantities related to the system process 20. The sensor 14 is designed to output a measured value 15 and the control unit 14 is designed to output a control command 17. Figure 1 The first embodiment shown assumes that the first step 110 has been completed, wherein the input signal array 30 is provided with a plurality of cells 32. Likewise, a monitoring data array 50 having a plurality of cells 52 is provided in the first step 110. The cells 52 of the monitoring data array 50 and the cells 32 of the input signal array 32 correspond to each other, so that the input data array 32 and the monitoring data array 52 are identical in structure.

[0031] In the first embodiment of the monitoring method 100, a second step 120 is performed, in which the system 10 is put into an active operating state, in which measured values ​​15 are provided by the sensor 14 and / or control commands 17 are provided for the input signal array 30 by the control unit 16. The provision of the measured values ​​15 or the control commands 17 is performed by storing them individually in the cells 32 of the input data array 30. The input signal array 30 is also provided 41 to a further processing unit 40, which is designed as an autoencoder 42. The third step 130 of the monitoring method 100 is performed by the further processing unit 40.

[0032] By means of a further processing unit 40, the content, i.e. the value, of a cell 52 of a monitoring data array 50 is determined based on the input signal array 30. Based on this, a comparison value 54 of the monitoring data array 50 is in turn determined, which value incorporates the content of its cell 52. Similarly, a comparison value 34 of the input signal array 30 is determined, which value also incorporates the content of its cell 32. The comparison values ​​34, 54 can be designed, for example, as a sum, respectively. In addition, by matching the comparison values ​​34, 54 of the input signal array 30 and the monitoring data array 50, a system deviation parameter 55 is determined. The system deviation parameter 55 is the total reconstruction error and represents the overall state of the system 10 in a compact form. The system deviation parameter 55 is compared with a system threshold value 57 in a third step 130. If the magnitude of the system deviation parameter 55 exceeds the system threshold value 57, an abnormal operating state of the system 10 is identified. The abnormal operating state identified in the third step 130 is non-specific, i.e. not suitable for indicating a potential cause of a defect. If an abnormal operating state of the system 10 is detected in the third step 130 , a fourth step 140 is carried out in the claimed monitoring method 100 .

[0033] In a fourth step 140, the content of the cell 52 of the monitoring data array 50 is compared independently of the content of the cell 32 of the input signal array 30, i.e. separately. Another processing unit 40 designed as an automatic encoder 42 maps the content of the cell 32 of the input signal array 30 to the content of the cell 52 of the monitoring data array 50 in a substantially identical manner during normal operation of the system 10. The mapping 43 is achieved by storing the content cell by cell. Therefore, in the fourth step 140, a cell deviation parameter 45 is determined for a plurality of cells 32 of the input signal array 30 and a corresponding cell 52 of the monitoring data array 50, respectively. For example, the cell deviation parameter 45 can be determined by the difference between the associated cells 32, 53. The cell deviation parameters 45 determined in this way are compared with the adjustable cell threshold values ​​47, respectively. This comparison is performed in a Figure 1 4 is symbolically indicated by an arrow 48. If the magnitude of the cell deviation parameter 45 exceeds the associated adjustable cell threshold 47, the associated cell 32, 52 of the input signal array 30 or the monitoring data array 50 is identified as a defective cell 58. For this purpose, the defective cell 58 is marked accordingly, for example with a so-called signature bit.

[0034] Furthermore, a fifth step 150 is performed in the monitoring method 100, in which the content of the cell 52 of the monitoring data array 50 is provided for matching with the database 60. Figure 1, the provision of a fifth step 150 is indicated by an arrow 41. In the fifth step 150, an overlimit 61 is determined for at least one defective cell 58, i.e. the content of the defective cell 58 exceeds the value of the associated cell deviation parameter 45. Furthermore, the cells 32, 52 of the monitoring data array 50 or the input signal array 30 are recorded, which exceed the value of the corresponding cell threshold value 47. From this, for example, in conjunction with the cell deviation parameter 45, whose value is below the associated cell threshold value 47, a pattern 62 is derived, which is characteristic of a potential cause of the defect. In the fifth step 150, this pattern 62 is identified by pattern matching 65 with the aid of a database 60. In the database 60, a plurality of patterns 62 are stored as comparison patterns. If the pattern 62 is identified in the fifth step 150, a corresponding warning 59 is issued. The warning 59 can indicate the presence of an abnormal operating state of the system 10, indicate a defective device 12 associated with the at least one defective cell 58, and / or indicate an identified cause of the defect. The warning 59 can be transmitted to a display unit 64, which is designed to issue the warning 59 to the user, for example in the form of a visual or acoustic warning. Alternatively or additionally, the warning 59 can be designed as an electronic warning signal, which is output to a communication interface 66, which can be connected to a higher-level control unit, not shown in detail. Figure 1 The monitoring method 100 is implemented by a computer program product 70 executed on a monitoring unit 75. The computer program product 70 forms an artificial intelligence 99 for monitoring a system 10 designed as a gas chromatograph 11.

[0035] Figure 2 A second embodiment of the claimed monitoring method 200 is schematically shown in FIG. The monitoring method 200 is used for monitoring a system 10 with which a system process 20 is executed. The system 10 is designed as a gas analysis device 11 and the system process 20 executed thereon comprises determining the composition 25 of a material sample 22. For this purpose, the system 10 is equipped with a plurality of devices 12, which are designed as sensors 14 or control units 14. The control unit 16 is designed to influence the system process 20 by executing a control program 18. The sensor 14 acquires physical quantities which are relevant to the system process 20. The sensor 14 is designed to output a measured value 15 and the control unit 14 is designed to output a control command 17.

[0036] Figure 2The second embodiment shown assumes that the first step 210 has been completed, wherein a neural network 80 is provided which has been trained for use in the monitoring method 200. The neural network 80 comprises a plurality of nodes 82 which are interconnected via edges 81. Each edge 81 is assigned a relevance value 83 which indicates to what extent a change at the node 82 (i.e. a measured value 15 or a control command 17 present there) follows the node 82 connected via the corresponding edge 81. The relevance threshold 83 can be adjusted, for example, by user input, a table of values ​​or an algorithm, in particular an artificial intelligence. According to Figure 2 , the edges 81 in the neural network 80 are mapped to zero correlation 85 in the normal operating state of the system 10. Therefore, the magnitude of the correlation value 83 of the corresponding edge 81 is below the adjustable correlation threshold 86, which basically describes a minimum interaction between the corresponding nodes 82. Therefore, a zero correlation 85 in the neural network 80 describes a measured value 15 or a control command 17, between which there is not even a significant correlation in the normal state. In a first step 210, the edges 81 corresponding to the zero correlation 85 are determined by training the neural network 80.

[0037] The monitoring method 200 comprises a second step 220, which is based on the completed first step 210. In the second step 220, the system 10 is placed in an active operating state, in which the system process 20 is executed and the device 12, i.e. the sensor 14 and the control unit 16, generates and provides the measured value 15 or the control command 17. Figure 2 15 or 26 provided in the second step 220, the correlation value 83 present at the respective edge 81 is determined, i.e. calculated. The existing correlation value 83 is stored, for example, in the cell 92 of the table 90 and a network deviation parameter 94 is calculated therefrom. The network deviation parameter 94 can be formed, for example, as the sum of the magnitudes of the correlation values ​​83. The network deviation parameter 94 is a measure of the presence of an unspecified abnormal operating state of the system 10. Furthermore, the monitoring method 200 comprises the execution of a fourth step 240. In the fourth step 240, the network deviation parameter 94 determined in the third step 230 is compared with a network threshold value 95. In the normal operating state of the system 10, the magnitude of the network deviation parameter 94 is below the network threshold value 95. The network threshold value 95 can be designed as a fixed threshold value or a moving threshold value, which can be increased, for example, depending on the operating time completed by the system 10. If an abnormal operating state is detected, a warning can be issued.

[0038] In addition, if Figure 2As shown, in the fifth step 250 of the monitoring method 200, the correlation value 83 of each edge 81 is compared with the associated correlation threshold value 86. If the magnitude of the correlation value 83 of the edge 81 exceeds the associated correlation threshold value 86, the associated edge 81 is identified as a defective edge 87. The defective edge 87 enables the inference that the measured value 15 or the control command 17 at one of the nodes 82 connected by the edge 81 is generated due to an abnormal operating state of the system 10. Therefore, it can be diagnosed that the device 12 belonging to the corresponding node 82 is defective or at least can be expected to be defective.

[0039] Furthermore, the monitoring method 200 comprises a sixth step 260, in which at least the correlation value 83 is provided for further evaluation. Figure 2 Indicated by arrows in the figure. In a sixth step 260, a pattern 62 is determined based on at least one identified defective edge 81, the pattern comprising a plurality of existing correlation values ​​83. To this end, the correlation values ​​83 of a plurality of edges 81 are compared jointly (i.e. substantially simultaneously) with respective associated correlation thresholds 86, which are zero correlations 85 in normal operating conditions. Based on the magnitude of the correlation values ​​83 and the extent to which these values ​​may exceed the associated correlation thresholds 86, a characteristic pattern 62 of the respective occurrence of defects can be derived. The pattern 62 is identified in the sixth step 260 by means of a database 60, which is used for pattern matching 65. Thus, the cause of the defect can be diagnosed by means of pattern matching 65. If a pattern 62 is identified in the sixth step 260, a corresponding warning 59 is issued. The warning 59 can indicate the presence of an abnormal operating state of the system 10, indicate a defective device 12 associated with the defective edge 81, and / or indicate an identified cause of the defect. The warning 59 can be transmitted to a display unit 64, which is designed to output the warning 59 to a user, for example in the form of a visual or audible warning. Alternatively or additionally, the warning 59 can be designed as an electronic warning signal which is output to a communication interface 66 which can be connected to a higher-level control unit (not shown in detail). Figure 2 The monitoring method 200 is implemented by a computer program product 70 executed on a monitoring unit 75. The computer program product 70 forms an artificial intelligence 99 for monitoring a system 10 designed as a gas chromatograph 11.

Claims

1. A monitoring method (200) for a system (10), the system having a plurality of devices (12), the devices being adapted to form nodes (82) in a neural network (80), the neural network mapping interactions of the devices (12) in the system (10), the monitoring method comprising the following steps: a) providing a trained neural network (80) having a plurality of edges (81), each of which corresponds to a zero correlation (85) or a running correlation, and in the case of the zero correlation (85) determining a correlation value (83) during training, which correlation value is below an adjustable correlation threshold value (86); b) providing the system (10) in an active operating state and acquiring measurement values ​​(15) and / or control commands (17) corresponding to at least one node (82) and / or one edge (81) of the neural network (80); c) determining a network deviation parameter (94), the network deviation parameter combining correlation values ​​(83) of a plurality of edges (81) of the neural network (80); wherein the correlation value (83) is determined using the measurement value (15) and / or the control command (17) obtained in step b); d) if the value of the network deviation parameter (94) exceeds an adjustable network threshold value (95), an abnormal operating state of the system (10) is identified; If the abnormal operating state is identified, a warning is issued (59). Characterized in that the monitoring method (200) comprises a further step e), in which, if the magnitude of the correlation value (83) of an edge (81) of the neural network (80) corresponding to a zero correlation (85) under normal operating conditions exceeds the associated correlation threshold value (86), the edge is identified as a defective edge (87).

2. The monitoring method (200) according to claim 1, characterized in that: Step b), step c) and / or step d) are performed at a first frequency for a running correlation and at a second frequency for a zero correlation (85).

3. The monitoring method (200) according to claim 1 or 2, characterized in that: If an abnormal state is detected in the first process, step b), step c) and / or step d) are performed in the first process for a running correlation and in the second process for a zero correlation (85).

4. The monitoring method (200) according to any one of claims 1 to 3, characterized in that: The relevance threshold (86) is designed as a fixed threshold or a moving threshold.

5. The monitoring method (200) according to any one of claims 1 to 4, characterized in that: The monitoring method (200) comprises a further step f), in which a pattern matching (65) is performed based on at least one identified defect edge (87) and a defect cause is identified.

6. A computer program product (70) for a monitoring system (10), the system having a plurality of devices (12) which interact to operate the system (10) and for which measured values ​​(15) and / or control commands (17) can be acquired and processed in a monitoring method (200), characterized in that The monitoring method (200) is designed as the monitoring method according to any one of claims 1 to 5.

7. A monitoring unit (75) for a system (10), the system comprising a plurality of devices (12) which interact to operate the system (10) and for which measured values ​​(15) and / or control commands (17) can be acquired, wherein: The monitoring unit (75) is designed to receive and process the measured values ​​(15) and / or the control commands (17), and the monitoring unit is designed to issue a warning (59), characterized in that the monitoring unit (75) is equipped with a computer program product (70) according to claim 6 for processing the measured values ​​(15) and / or the control commands (17).

8. A gas analysis device (11) comprising a plurality of devices (12) for conditioning and measuring a material sample (22), the gas analysis device being provided with a monitoring unit (75) for identifying an abnormal operating state of the gas analysis device (11), characterized in that: The monitoring unit (75) is designed as a monitoring unit according to claim 7.

9. An application of artificial intelligence (99) for monitoring a gas analysis device (11), wherein: The application at least includes the issuance of a warning (59), and the artificial intelligence (99) executes a monitoring method (200), characterized in that the monitoring method (100, 200) is designed as a monitoring method according to any one of claims 1 to 5.

Images