Anomaly detection and processing method for intelligent factory production process

Abstract

The application belongs to the technical field of lean manufacturing monitoring, and particularly relates to an abnormality detection and processing method for an intelligent factory production process, which comprises the following steps: based on a preset production process flow chain, a double-monitoring node chain and a layered abnormality correlation evaluation node chain are constructed, and equipment operation state and process parameter data are collected in real time; the layered abnormality correlation evaluation node chain analyzes the equipment health degree through an operation state evaluation sub-chain, and mines cross-process abnormality propagation relationships through a whole-process process abnormality evaluation sub-chain, generates a whole-process correlation abnormality evaluation space, simultaneously fuses real-time data and historical abnormality strategy index information based on a distributed algorithm, dynamically constructs a differential abnormality processing strategy library, and realizes synchronous feedback of the processing strategy for locating the abnormality; through the double-monitoring architecture and layered correlation evaluation analysis, the application solves the problems of fuzzy abnormality propagation path and delayed response, and improves the real-time performance and processing accuracy of abnormality detection.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of lean manufacturing monitoring, and particularly relates to an abnormality detection and processing method for a production process of an intelligent factory. BACKGROUND

[0002] With the improvement of the quality requirements of high-end manufacturing industry on high-precision steel pipes and pipe fittings, the production process based on lean manufacturing needs to realize fine control in the whole life cycle. In seamless pipe fitting production, the complex process covering more than ten procedures such as blanking, pickling and solid solution, and the slight deviation of process parameters in each link (such as acid concentration fluctuation and size tolerance overrun) may cause cross-process chain abnormality, resulting in batch rejection or production line paralysis. However, the existing abnormality detection technology has systematic defects: firstly, isolated detection and lag response, relying on single-point sensors or manual sampling, lacking of data cooperation between procedures, for example, the abnormality of the pickling procedure is only identified in the subsequent shot blasting or finished product inspection stage, resulting in that the abnormality cannot be found and intervened in time; secondly, difficult positioning, the abnormality source tracing relies on manual experience judgment, and the specific influence relationship of the abnormality of the previous procedure on the subsequent link (such as how the parameter deviation of the forming procedure is transmitted to the shot blasting defect) cannot be quantified, resulting in low fault troubleshooting efficiency; thirdly, static threshold limitation, the fixed alarm rule cannot adapt to dynamic production conditions (such as raw material batch difference and equipment performance degradation), resulting in frequent false alarm or missed detection, affecting the production continuity. Therefore, the application provides an abnormality detection and processing method for a production process of an intelligent factory. SUMMARY

[0003] In view of the defects of the prior art, the application provides an abnormality detection and processing method for a production process of an intelligent factory, which comprises the following steps: based on a preset production process flow chain, a double-monitoring node chain and a hierarchical abnormality correlation evaluation node chain are constructed, and equipment running state and process parameter data are collected in real time; the hierarchical abnormality correlation evaluation node chain analyzes the equipment health degree through a running state evaluation sub-chain, and digs the cross-process abnormality propagation relationship through a whole-process process abnormality evaluation sub-chain, generates a whole-process correlation abnormality evaluation space, and simultaneously based on a distributed algorithm, fuses real-time data and historical abnormality strategy index information, dynamically constructs a differential abnormality processing strategy library, and realizes synchronous feedback of the processing strategy of the located abnormality; through the double-monitoring architecture and hierarchical correlation analysis, the application breaks through the limitation of traditional isolated detection, solves the problems of fuzzy abnormality propagation path and lag response, and improves the real-time performance and processing accuracy of abnormality detection.

[0004] To achieve the above-mentioned purpose, the application provides the following technical scheme:

[0005] An abnormality detection and processing method for a production process of an intelligent factory, comprising:

[0006] Based on the preset production process chain, a dual monitoring node chain and a hierarchical abnormal association evaluation node chain are constructed;

[0007] The hierarchical abnormal association evaluation node chain includes a running state evaluation sub-chain and a whole-process process abnormality evaluation sub-chain;

[0008] Based on the dual monitoring node chain, the production process chain is monitored in real time to obtain device running state information and to-be-processed device production process information;

[0009] Based on the device running state information and the to-be-processed device production process information in combination with the hierarchical abnormal association evaluation node chain, a whole-process associated abnormality evaluation space is obtained;

[0010] Based on the whole-process associated abnormality evaluation space, a differential abnormality processing strategy is obtained through a distributed algorithm in combination with a historical abnormality strategy index information constructed abnormality processing strategy library, and the differential abnormality processing strategy is fed back to the dual monitoring node chain in real time, so as to realize real-time early warning and processing of the monitored and positioned abnormality in the production process chain.

[0011] Specifically, the production process chain includes a process node sub-chain and a device running node sub-chain; the dual monitoring node chain includes a device running monitoring sub-chain and a process monitoring sub-chain;

[0012] Each processing process node in the process node sub-chain saves the processing process parameters of the corresponding node; each running node in the device running node sub-chain saves the device control parameters of the corresponding processing device for executing the processing process;

[0013] The device running monitoring sub-chain and the device running node sub-chain are one-to-one corresponding; the process node sub-chain and the process monitoring sub-chain are one-to-one corresponding;

[0014] Each monitoring node in the device running monitoring sub-chain corresponds to a running evaluation node in the running state evaluation sub-chain.

[0015] Specifically, each process monitoring node in the process monitoring sub-chain corresponds to each process evaluation node in the whole-process process abnormality evaluation sub-chain;

[0016] Each running evaluation node in the running state evaluation sub-chain is connected to the process evaluation nodes in the whole-process process abnormality evaluation sub-chain in one-to-many backward mapping;

[0017] Each running evaluation node is configured with a running hierarchical evaluation index set for evaluating the running state of the corresponding processing device; each process evaluation node is configured with a hierarchical process evaluation index set for evaluating the corresponding processing process;

[0018] The operation hierarchical evaluation index set and the hierarchical process evaluation index set are respectively stored in a preset hierarchical operation index library and a hierarchical process index library, and the hierarchical process index library and the hierarchical operation index library are connected in one-to-many direction by a preset directional association connection.

[0019] Specifically, the construction process of the production process chain includes:

[0020] Obtain the full-process production process information of each device to be processed, and combine the pre-trained production process analysis algorithm to obtain the continuous processing process node sequence and processing order information corresponding to each device to be processed, as well as the required processing equipment sequence and processing equipment operation interaction information, and the mapping information and correlation causal information of each processing process and processing equipment;

[0021] Based on the continuous processing process node sequence and processing order information corresponding to each device to be processed, a loose coupling algorithm is used to construct the directional loose coupling connection between each processing process node corresponding to each device to be processed;

[0022] Based on the continuous processing process node sequence and the directional loose coupling connection, the process node sub-chain corresponding to each device to be processed is obtained;

[0023] Meanwhile, based on the required processing equipment sequence, the equipment operation node sequence is constructed, and the bidirectional operation abnormality reasoning connection is constructed by using the graph neural network and the processing equipment operation interaction information;

[0024] Based on the equipment operation node sequence and the bidirectional operation abnormality reasoning connection, the equipment operation node sub-chain is obtained.

[0025] Specifically, the construction process of the production process chain further includes:

[0026] Suppose that the current process node sub-chain contains M continuous processing process nodes, the equipment operation node sub-chain contains N equipment operation nodes, and the process parameters corresponding to each processing process node are only processed and operated on one equipment operation node;

[0027] Based on the assumption, the processing process node corresponding to the process parameters operated on each equipment operation node is set as the main processing process node of the equipment operation node, and the remaining continuous processing process nodes after the main processing process node constitute the causal correlation process node sequence of the current equipment operation node;

[0028] Based on the main causal correlation information between the operation state and the processing process result between each equipment operation node and the corresponding main processing process node, a self-correlation algorithm is used to construct the main correlation degree between each equipment operation node and the corresponding main processing process node, and the main connection relationship between the corresponding equipment operation node and the main processing process node is constructed by using the obtained main correlation degree.

[0029] Meanwhile, based on the cooperative operation causal information between each device operation node and the corresponding causal correlation process node sequence, the cooperative causal correlation degree of each device operation node to each processing process node of the corresponding causal correlation process node sequence is obtained through a cross-correlation algorithm.

[0030] Specifically, the construction process of the production process flow chain further includes:

[0031] Based on the cooperative causal correlation degree of each device operation node to each processing process node of the corresponding causal correlation process node sequence, a cooperative causal process connection between the device operation node and the corresponding causal correlation process node sequence is constructed;

[0032] Based on the main connection relationship and the cooperative causal process connection, the nodes corresponding to the process node sub-chain and the device operation node sub-chain are connected to obtain a production process flow chain;

[0033] Based on a Bayesian algorithm, a causal reasoning model is constructed and configured into the bidirectional operation abnormality reasoning connection and the cooperative causal process connection to obtain a production process flow chain with causal reasoning.

[0034] Specifically, the construction process of the dual monitoring node chain includes:

[0035] Based on the architecture mapping of the device operation node sub-chain, a device operation monitoring sub-chain with the same architecture is obtained, and a monitoring device is configured according to the corresponding operation attribute of each device operation node to obtain a device operation parameter space corresponding to the device operation node sub-chain;

[0036] The device operation parameter space includes a single operation parameter sequence corresponding to each device operation node and an interactive operation parameter sequence corresponding between the device operation nodes;

[0037] Meanwhile, based on the architecture mapping of the process node sub-chain, a process monitoring sub-chain with the same architecture is obtained, and a corresponding monitoring device is configured according to the processing process information to obtain a processing process monitoring information space corresponding to the device operation node sub-chain.

[0038] Specifically, the construction process of the hierarchical abnormality correlation evaluation node chain includes:

[0039] Based on the architecture mapping of the device operation node sub-chain, a running state evaluation sub-chain with the same architecture is obtained, and based on the architecture mapping of the process node sub-chain, a whole-process process abnormality evaluation sub-chain with the same architecture is obtained;

[0040] Map the main connection relationship in the production process chain with causal reasoning to the corresponding nodes between the running state evaluation sub-chain and the whole-process process abnormality evaluation sub-chain, and a one-to-many backward mapping connection between the running state evaluation sub-chain and the whole-process process abnormality evaluation sub-chain is constructed to obtain a hierarchical abnormality correlation evaluation node chain.

[0041] Specifically, the acquisition process of the differential abnormality processing strategy includes:

[0042] Based on the dual monitoring node chain, the equipment running parameter space and the processing process monitoring information space are monitored and acquired in real time, and are synchronously mapped to the corresponding running state evaluation sub-chain and whole-process process abnormality evaluation sub-chain in the hierarchical abnormality correlation evaluation node chain;

[0043] Based on the running hierarchical evaluation index set configured in each running evaluation node and the hierarchical process evaluation index set configured in each process evaluation node in the running state evaluation sub-chain and the whole-process process abnormality evaluation sub-chain, the running state and the processing process result of each running evaluation node and the corresponding main processing process node are synchronously evaluated to obtain a running state evaluation result sequence and a processing process evaluation result sequence;

[0044] A running state abnormality evaluation threshold sequence and a processing process abnormality evaluation threshold sequence are set, and based on the running state evaluation result sequence and the processing process evaluation result sequence, the running state abnormality evaluation threshold sequence and the processing process abnormality evaluation threshold sequence are combined to synchronously discriminate the equipment running node sub-chain and the process node sub-chain, and a whole-process correlation abnormality evaluation space is obtained.

[0045] Compared with the prior art, the beneficial effects of the present application are:

[0046] The present application is aimed at the deficiencies of the prior art, and by constructing a dual monitoring node chain and a hierarchical abnormality correlation evaluation node chain, full-dimensional real-time synchronous monitoring of equipment running state and processing process information is realized, multi-level threshold discrimination of running hierarchical evaluation index set and hierarchical process evaluation index set is combined, and complex scenarios such as single equipment abnormality, process independent abnormality and cross-node collaborative abnormality are effectively distinguished; based on the Bayesian causal reasoning model and the collaborative causal process connection, the abnormality propagation path and the correlation contribution degree of the equipment running node and the process node are accurately located, and the depth and accuracy of abnormality tracing are strengthened; by means of a distributed algorithm, a historical abnormality strategy library is dynamically matched, differential processing strategies are generated for different abnormality types, bidirectional running abnormality reasoning connection and process parameter deviation reverse deduction are combined, closed-loop dynamic compensation correction of equipment control parameters and process parameters is realized, real-time and self-adaptability of abnormality processing is improved, and the risk of global process chain fluctuation caused by local abnormality is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 Flowchart of seamless pipe fitting process for Embodiment 1 of the present application;

[0048] Figure 2 Flowchart of an abnormality detection and processing method for a smart factory production process for Embodiment 1 of the present application;

[0049] Figure 3 Logic architecture diagram of an abnormality detection and processing method for a smart factory production process for Embodiment 1 of the present application;

[0050] Figure 4 Abnormality monitoring architecture diagram of a production process flow chain for Embodiment 1 of the present application. DETAILED DESCRIPTION

[0051] Embodiment 1

[0052] With the increasing requirements of high-end manufacturing on the quality of high-precision steel pipes and pipe fittings, the production process based on lean manufacturing needs to realize fine control throughout the life cycle. In the production and control of fittings, such as the seamless pipe fitting process in Figure 1 , which covers more than ten complex processes such as blanking, pickling, and solid solution, small deviations in process parameters (such as fluctuations in acid concentration and size tolerance) at each link may cause cross-process chain abnormalities, resulting in batch rejection or production line paralysis. However, existing abnormality detection techniques have problems such as isolated detection and lagging response, difficult positioning, and limitations of relying on manual experience judgment and static threshold for abnormal root cause tracing, which all lead to frequent false positives or missed detections, affecting production continuity, etc. Therefore, referring to Figure 2 , Figure 3 , the present application provides an embodiment: an abnormality detection and processing method for a smart factory production process, comprising the steps of:

[0053] S1, based on a preset production process flow chain, constructing a dual-monitoring node chain and a hierarchical abnormality correlation evaluation node chain;

[0054] Further, the hierarchical abnormality correlation evaluation node chain in this embodiment includes a running state evaluation sub-chain and a whole-process process abnormality evaluation sub-chain;

[0055] The production process flow chain includes a process node sub-chain and a device running node sub-chain; the dual-monitoring node chain includes a device running monitoring sub-chain and a process monitoring sub-chain;

[0056] Further, each processing process node in the process node sub-chain in this embodiment saves the processing process parameters of the corresponding node;

[0057] Further, the specific processing parameters in the embodiment are set by a person skilled in the art according to the specific processing procedure; the device control parameters of the device running node sub-chain are also the same, and are set by a person skilled in the art according to the machine type and the specific device;

[0058] Each running node in the device running node sub-chain saves the device control parameters for the corresponding processing device to execute the processing technology;

[0059] The device running monitoring sub-chain and the nodes in the device running node sub-chain correspond one by one; the process node sub-chain and the nodes in the process monitoring sub-chain correspond one by one;

[0060] Each monitoring node in the device running monitoring sub-chain corresponds to a running evaluation node in the running state evaluation sub-chain one by one.

[0061] Further, each process monitoring node in the process monitoring sub-chain in the embodiment corresponds to each process evaluation node in the whole-process process abnormality evaluation sub-chain one by one;

[0062] Each running evaluation node in the running state evaluation sub-chain is connected to the process evaluation nodes in the whole-process process abnormality evaluation sub-chain in a one-to-many backward mapping manner;

[0063] Each running evaluation node is configured with a running hierarchical evaluation index set for evaluating the running state of the corresponding processing device; each process evaluation node is configured with a hierarchical process evaluation index set for evaluating the corresponding processing technology;

[0064] The running hierarchical evaluation index set and the hierarchical process evaluation index set are respectively stored in a preset hierarchical running index library and a hierarchical process index library, and the hierarchical process index library and the hierarchical running index library are connected in a one-to-many directional connection manner through a preset directional association connection.

[0065] Further, the specific process of connecting the hierarchical process index library and the hierarchical running index library in a one-to-many directional connection manner through a preset directional association connection in the embodiment includes:

[0066] According to all the hierarchical process evaluation index sets and all the running hierarchical evaluation index sets, the correlation influence degree of each device running node corresponding running hierarchical evaluation index and the corresponding hierarchical process evaluation index in the subsequent processing technology node is obtained through correlation analysis method analysis, and a one-to-many directional connection between each running hierarchical evaluation index and multiple hierarchical process evaluation indexes is constructed based on the correlation influence degree of the device running node corresponding running hierarchical evaluation index and the corresponding hierarchical process evaluation index in the subsequent processing technology node.

[0067] Further, in the seamless steel pipe production process, in order to realize accurate control of the production process and quality optimization, the operation hierarchical evaluation index set and the hierarchical process evaluation index set are respectively stored in the preset hierarchical operation index library and the hierarchical process index library, and are connected in one-to-many direction through the preset directional association connection;

[0068] Further, in order to better illustrate the process node sub-chain construction process in this embodiment, taking the seamless pipe production process as an example, the process node sub-chain includes more than ten processing process nodes corresponding to processes such as blanking, pickling, and solid solution. Each processing process node saves the processing process parameters of the corresponding process. For example, in the pickling process, parameters such as acid concentration, pickling time, and temperature are recorded. In the blanking process, the size specification and cutting accuracy of the pipe are key process parameters. These parameters are the basis for ensuring the normal operation of each process and guaranteeing product quality. They determine the specific operation standards of each process.

[0069] In the device operation node sub-chain construction process: each operation node in the device operation node sub-chain corresponds to a device that executes each processing process. Each operation node saves the device control parameters of the corresponding processing device executing the processing process. For example, in the acid pickling device operation node, the stirring speed and acid circulation flow rate of the device are recorded and stored. For the solid solution treatment device, parameters such as heating temperature control, heating time control, and cooling rate control are information that the device operation node needs to save. These parameters directly affect the running state of the device and thus affect the execution effect of the process.

[0070] Further, the construction process of the dual monitoring node chain in this embodiment includes:

[0071] The device operation monitoring sub-chain in this embodiment corresponds one-to-one to the nodes in the device operation node sub-chain. Taking the seamless pipe production as an example, various sensors such as temperature sensors, pressure sensors, and flow sensors are installed on each device. For the acid pickling device, the temperature sensor monitors the acid temperature, the pressure sensor monitors the pressure of the acid circulation system, and the flow sensor monitors the flow of the acid. These sensors collect various data during the device operation process in real time, forming the data source of the device operation monitoring sub-chain. Through real-time monitoring of these data, abnormal conditions in the device operation process can be discovered in a timely manner. For example, excessively high temperature may indicate a problem with device heat dissipation, and abnormal pressure fluctuations may indicate pipe blockage or pump malfunction.

[0072] Further, each process monitoring node in the process monitoring sub-chain in the embodiment corresponds to a node in the process node sub-chain; monitoring points are set at each process link to monitor key indicators of different processes; in the pickling process, in addition to monitoring the temperature, concentration and other equipment operation related parameters of the acid solution, the corrosion degree and cleanliness of the pipe surface and other process indicators also need to be monitored; in the cutting process, the size of the cut pipe is measured in real time to monitor whether it meets the specified size tolerance range. These process monitoring data can directly reflect the effect of process execution, helping to determine whether the process is within the normal range. It needs to be further explained that the monitoring points are set at each process link in the embodiment, and the monitoring equipment used by the monitoring points is set by a person skilled in the art according to the monitoring process requirements; for example, in the cutting process, the size of the cut pipe is monitored by using a laser diameter measuring instrument to scan the outer diameter of the pipe, combined with a contact grating ruler to measure the length, and through a data processing system to calculate in real time whether the size deviation is within the tolerance range; for the perpendicularity and other form and position tolerances of the pipe, a three-coordinate measuring machine is used for sampling detection, or a vision detection system is used to shoot the pipe end surface image to analyze the edge perpendicularity deviation.

[0073] Further, each monitoring node in the equipment operation monitoring sub-chain in the embodiment corresponds to a running evaluation node in the running evaluation sub-chain; the running evaluation node receives monitoring data from the equipment operation monitoring sub-chain, and uses a comprehensive fuzzy algorithm combined with a Bayesian function to evaluate the running state of the equipment; for example, a parameter model under the normal running state of the equipment is established by using historical data and a comprehensive fuzzy algorithm combined with a Bayesian function, when the monitoring data deviates from the normal model range, the evaluation node can determine that the equipment is in an abnormal state, and for the pickling equipment, if the acid solution temperature continuously exceeds the normal range and the fluctuation amplitude exceeds the set threshold, the running evaluation node will determine that the equipment running state is abnormal. At the same time, combined with the running time, maintenance record and other information of the equipment, the running condition of the equipment can be more accurately evaluated, and possible faults can be predicted.

[0074] Further, each process monitoring node in the process monitoring sub-chain in the embodiment corresponds to each process evaluation node in the whole-process process abnormality evaluation sub-chain; the whole-process process abnormality evaluation sub-chain comprehensively analyzes the process data transmitted by each process monitoring node, combined with the requirements and standards of the entire production process, to determine whether the process is abnormal; for example, in seamless pipe production, not only the process parameters of a single process need to be considered, but also the correlation between processes. If the cleanliness of the pipe surface after pickling does not meet the standard, it may affect the effect of subsequent solid solution treatment, and thus affect the product quality. By comprehensively analyzing the data of multiple process monitoring points, whether the process is normal can be more comprehensively determined.

[0075] The connection between the running state evaluation sub-chain and the whole-process process abnormality evaluation sub-chain: each running evaluation node in the running state evaluation sub-chain is connected to the process evaluation nodes in the whole-process process abnormality evaluation sub-chain in a one-to-many backward mapping manner, which means that when the whole-process process abnormality evaluation sub-chain finds that an abnormality occurs in a certain process link, the running state information related to the process link can be quickly located through the mapping relationship; for example, if it is found that the quality of the pipe after the solution treatment is problematic, the whole-process process abnormality evaluation sub-chain can check the running state evaluation information of the solution treatment equipment and the equipment of the previous related processes through the mapping relationship;

[0076] S2, real-time monitoring of the production process chain based on the double monitoring node chain to obtain device running state information and to-be-processed device production process information;

[0077] S3, obtaining a whole-process associated abnormality evaluation space based on the device running state information and the to-be-processed device production process information in combination with the hierarchical abnormality association evaluation node chain;

[0078] S4, obtaining a differential abnormality processing strategy based on the whole-process associated abnormality evaluation space and an abnormality processing strategy library constructed by a distributed algorithm in combination with historical abnormality strategy index information, and feeding back the differential abnormality processing strategy to the double monitoring node chain in real time to perform real-time early warning and processing on the monitored and located abnormality in the production process chain.

[0079] Further, the abnormality processing strategy library in the embodiment is connected by the corresponding association degree between each abnormality strategy keyword in the abnormality processing strategy library and the running index keyword in the hierarchical running index library.

[0080] Further, the abnormality processing strategy library in the embodiment quickly indexes the associated abnormality processing strategies in the hierarchical running index library by an association analysis algorithm in combination with the association mapping connection to obtain a differential abnormality processing strategy.

[0081] Further, the construction process of the production process chain in the embodiment includes:

[0082] Obtaining the whole-process production process information of each to-be-processed device in combination with a pre-trained production process analysis algorithm to obtain the corresponding continuous processing process node sequence and processing sequence information of each to-be-processed device, the required processing equipment sequence and processing equipment running interaction information, and the mapping information and associated causal information of each processing process and processing equipment;

[0083] Further, the production process analysis algorithm in the embodiment is constructed by a pre-trained Chinese Bert model.

[0084] Based on the continuous processing process node sequence and the processing sequence information corresponding to each device to be processed, a loose coupling algorithm is used to construct a directed loose coupling connection between each processing process node corresponding to each device to be processed;

[0085] Based on the continuous processing process node sequence and the processing sequence information corresponding to each device to be processed, a loose coupling algorithm is used to construct a directed loose coupling connection between each processing process node corresponding to each device to be processed;

[0086] Meanwhile, based on the required processing equipment sequence, an equipment operation node sequence is constructed, and a bidirectional operation abnormality reasoning connection is constructed through a graph neural network using the processing equipment operation interaction information;

[0087] Based on the equipment operation node sequence and the bidirectional operation abnormality reasoning connection, an equipment operation node sub-chain is obtained.

[0088] Suppose that the current process node sub-chain contains M continuous processing process nodes, the equipment operation node sub-chain contains N equipment operation nodes, and the process parameters corresponding to each processing process node are only processed and operated on one equipment operation node;

[0089] Based on the assumption, the processing process node corresponding to the process parameter operated on each equipment operation node is set as the main processing process node of the equipment operation node, and the remaining continuous processing process nodes after the main processing process node constitute the causal correlation process node sequence of the current equipment operation node;

[0090] Based on the main causal correlation information between each equipment operation node and the corresponding main processing process node in the running state and the processing process result, a self-correlation algorithm is used to construct the main correlation degree between each equipment operation node and the corresponding main processing process node, and the main connection relationship between the corresponding equipment operation node and the main processing process node is constructed using the obtained main correlation degree;

[0091] Meanwhile, based on the cooperative running causal information between each equipment operation node and the corresponding causal correlation process node sequence, a cross-correlation algorithm is used to obtain the cooperative causal correlation degree of each equipment operation node for each processing process node in the corresponding causal correlation process node sequence;

[0092] Based on the cooperative causal correlation degree of each equipment operation node for each processing process node in the corresponding causal correlation process node sequence, a cooperative causal process connection between the equipment operation node and the corresponding causal correlation process node sequence is constructed;

[0093] Based on the main connection relationship and the cooperative causal process connection, the nodes corresponding to the process node sub-chain and the equipment operation node sub-chain are connected to obtain a production process flow chain;

[0094] For a more detailed description of the construction process of the production process flow chain, please refer toFigure 4 Wherein A represents a process node sub-chain, B represents a device running node sub-chain, the process node sub-chain contains M continuous processing process nodes, the device running node sub-chain contains N device running nodes, the corresponding single arrow in A is a directed loose coupling connection, which is used for safe transmission and interaction of process parameters by using blockchain technology; the double-headed arrow in B is a bidirectional running abnormal reasoning connection, which is used for forward or backward abnormal tracing of abnormalities or faults, and positioning of abnormalities or faults.

[0095] Further, in the embodiment, 1 to M-1 between A and B are main connection relationships and collaborative causal process connections between the device running node 1 and the processing process nodes 1 to M, wherein in the corresponding connection between A and B, 1 represents the main connection, and 2 to M-1 are collaborative causal process connections of the device running node 1 and the corresponding causal associated process node sequence, which are used for forward tracing of processing equipment production process deviation when processing is abnormal and the device does not detect abnormality, and accurate correction of deviation on the accumulated path of the traced process deviation.

[0096] A causal reasoning model is constructed based on a Bayesian algorithm and configured into the bidirectional running abnormal reasoning connection and the collaborative causal process connection, so as to obtain a production process flow chain with causal reasoning.

[0097] Further, the construction process of the double monitoring node chain in the embodiment includes:

[0098] A device running monitoring sub-chain with the same architecture is obtained based on the architecture mapping of the device running node sub-chain, and a monitoring device is configured according to the corresponding running attribute of each device running node, so as to obtain a device running parameter space corresponding to the device running node sub-chain;

[0099] The device running parameter space includes a single running parameter sequence corresponding to each device running node and an interactive running parameter sequence corresponding between the device running nodes;

[0100] Meanwhile, a process monitoring sub-chain with the same architecture is obtained based on the architecture mapping of the process node sub-chain, and a corresponding monitoring device is configured according to the processing process information, so as to obtain a processing process monitoring information space corresponding to the device running node sub-chain.

[0101] Further, the construction process of the hierarchical abnormal association evaluation node chain in the embodiment includes:

[0102] The running state evaluation sub-chain with the same architecture is obtained based on the architecture mapping of the device running node sub-chain, and the whole-process process abnormality evaluation sub-chain with the same architecture is obtained based on the architecture mapping of the process node sub-chain;

[0103] The main connection relationship in the production process chain with causal reasoning is mapped between the corresponding nodes in the operation state evaluation sub-chain and the whole-process process abnormality evaluation sub-chain, and a one-to-many backward mapping connection between the operation state evaluation sub-chain and the whole-process process abnormality evaluation sub-chain is constructed to obtain a hierarchical abnormality correlation evaluation node chain.

[0104] Further, the difference abnormality processing strategy acquisition process in the embodiment includes:

[0105] Based on the dual monitoring node chain, the equipment operation parameter space and the processing technology monitoring information space are monitored and acquired in real time, and are synchronously mapped to the corresponding operation state evaluation sub-chain and the whole-process process abnormality evaluation sub-chain in the hierarchical abnormality correlation evaluation node chain;

[0106] Based on the operation hierarchical evaluation index set configured in each operation evaluation node in the operation state evaluation sub-chain and the whole-process process abnormality evaluation sub-chain and the hierarchical process evaluation index set configured in each process evaluation node, the operation state and the processing technology result of each operation evaluation node and the corresponding main processing technology node are synchronously evaluated to obtain an operation state evaluation result sequence and a processing technology evaluation result sequence;

[0107] An operation state abnormality evaluation threshold sequence and a processing technology abnormality evaluation threshold sequence are set, and based on the operation state evaluation result sequence and the processing technology evaluation result sequence, the operation state abnormality evaluation threshold sequence and the processing technology abnormality evaluation threshold sequence are combined to synchronously discriminate the equipment operation node sub-chain and the process node sub-chain to obtain a whole-process correlation abnormality evaluation space.

[0108] Based on the whole-process correlation abnormality evaluation space, if at least one pair of equipment operation node and corresponding main processing technology node in the equipment operation node sub-chain and the process node sub-chain is in an abnormal state, a first difference abnormality processing strategy is obtained according to the abnormality evaluation score of the located abnormal equipment operation node and the corresponding main processing technology node, the processing technology parameter and the equipment control parameter, and the abnormality processing strategy library.

[0109] It needs to be further explained that one implementation manner of the first difference abnormality processing strategy acquisition in the embodiment is:

[0110] Based on the whole-process correlation abnormality evaluation space, the equipment operation node and the corresponding main processing technology node in the abnormal state are located, and the abnormality evaluation score, the processing technology parameter deviation value and the equipment control parameter deviation value of the abnormal node are extracted.

[0111] An operation index keyword is extracted from the anomaly evaluation score, the processing process parameter deviation value and the equipment control parameter deviation value by an entity-relation extraction algorithm, and the operation index keyword includes an equipment type, a process type, a parameter name and a deviation range;

[0112] The extracted operation index keyword is matched with a preset abnormal strategy keyword in an abnormal processing strategy library, and a candidate abnormal processing strategy with a correlation degree greater than or equal to a threshold value is screened out by a correlation analysis algorithm;

[0113] It should be further explained that the abnormal processing strategy library in the embodiment pre-stores a plurality of abnormal processing strategy records, and each record contains a strategy ID and an abnormal strategy keyword;

[0114] The anomaly evaluation score of the anomaly node is classified according to a preset anomaly level interval, the candidate abnormal processing strategies are prioritized according to the classification result, the parameters of the sorted candidate strategies are adjusted according to the equipment operation history maintenance record and the process execution history data, and a first difference abnormal processing strategy is generated;

[0115] It should be further explained that the embodiment adjusts the parameters in the candidate strategy in combination with the operation and maintenance records and process execution history data of the same type of equipment within a preset time length, for example, the optimal correction value under the same deviation;

[0116] It should be further explained that the first difference abnormal processing strategy generated in the embodiment needs to include but is not limited to: target node ID, correction parameter name, parameter value before and after correction, execution time limit, etc.;

[0117] The first difference abnormal processing strategy is fed back to the double monitoring node chain in real time, triggers the corresponding abnormal early warning mechanism, and drives the equipment to perform parameter correction operation.

[0118] If only the processing process node is in an abnormal state at the current time, and the equipment operation node is evaluated as normal, then according to the corresponding equipment operation node of the abnormal processing process node, the deviation accumulation equipment operation node sequence is obtained by performing processing abnormal double forward reasoning positioning through the bidirectional operation abnormal reasoning connection and the collaborative causal process connection built in the production process flow chain with causal reasoning;

[0119] Based on the deviation accumulation equipment operation node, a processing process evaluation result corresponding to the main processing process node is obtained;

[0120] Based on the processing process evaluation result corresponding to the main processing process node, a process parameter deviation value satisfying the process requirement corresponding to each main processing process node is obtained;

[0121] obtain the equipment control parameter deviation value of the equipment running node corresponding to each main machining process node based on the process parameter deviation value meeting the process requirement corresponding to each main machining process node;

[0122] obtain the second difference abnormality processing strategy based on the equipment control parameter deviation value of the deviation accumulation equipment running node and in combination with the abnormality processing strategy library, and correct the process parameter deviation value of the deviation accumulation equipment running node based on the second difference abnormality processing strategy, so that the machining process evaluation result corresponding to each machining process node meets the corresponding machining process abnormality evaluation threshold in real time.

[0123] It should be further explained that the implementation mode of the second difference abnormality processing strategy in the embodiment is specifically as follows:

[0124] Based on the unique identification ID of the abnormal process node, the corresponding equipment running node is determined, and the bidirectional running abnormality inference connection and the collaborative causal process connection of the built-in Bayesian causal inference model in the production process flow chain are used to perform forward inference starting from the abnormal process node, calculate the deviation transmission coefficient of the equipment running node to the subsequent process node, and select the equipment running node with the deviation transmission coefficient ≥ the preset threshold, and arrange the deviation accumulation equipment running node sequence in descending order according to the deviation influence weight, wherein the deviation transmission coefficient is obtained by mutual correlation analysis of historical process deviation data and equipment control parameters.

[0125] Based on the main machining process node identifier of each node in the deviation accumulation equipment running node sequence, the corresponding process evaluation result in the whole-process process abnormality evaluation sub-chain is called, the difference between the current process parameter measured value and the process standard value of the main machining process node is extracted, the process parameter deviation value meeting the process requirement is calculated by combining the allowed deviation threshold of the main machining process node in the hierarchical process index library and the deviation compensation algorithm, and the parameter sensitivity coefficient of the subsequent causal correlation process node needs to be considered to ensure that the calculation result is within the process tolerance interval.

[0126] According to the preset parameter mapping table in the main connection relationship between the process node sub-chain and the equipment running node sub-chain, the determined process parameter deviation value is converted into the equipment control parameter deviation value of the corresponding equipment running node, and the conversion process needs to introduce the equipment characteristic curve correction coefficient, which is obtained by comparing and analyzing the factory parameters and the actual running calibration data of the equipment, for example, when the acid concentration deviation value of the pickling process is converted into the acid liquid circulation flow deviation value of the pickling equipment, the dynamic correction needs to be performed according to the acid concentration-flow response curve.

[0127] The obtained equipment control parameter deviation value is taken as a keyword, cosine similarity matching is performed between the keyword and an abnormal strategy keyword stored in an abnormal processing strategy library, a candidate strategy with the highest similarity is screened out, a control parameter adjustment range in the candidate strategy is adapted and optimized in combination with a model parameter of a deviation accumulation equipment running node and a historical correction record, a second difference abnormal processing strategy including a correction object ID, a control parameter name, a value before and after adjustment, an execution order and a verification threshold is generated, the strategy is issued to a corresponding equipment running node through a double monitoring node chain, the equipment is driven to perform a parameter correction operation, and a corrected process parameter is collected in real time through a process monitoring sub-chain until a process evaluation result meets a corresponding processing process abnormal evaluation threshold.

[0128] In summary, the embodiment realizes multi-dimensional abnormal monitoring and accurate regulation system in seamless steel pipe production scene by constructing the double-chain collaborative architecture of process node sub-chain and equipment running node sub-chain, combining the dynamic perception of double monitoring node chain and the causal reasoning mechanism of hierarchical abnormal association evaluation node chain. Specifically, first, the equipment running monitoring sub-chain captures the high-frequency time sequence characteristics of equipment control parameters in real time through a multi-dimensional sensor network, and establishes an equipment health baseline model with the aid of the Bayesian-fuzzy comprehensive algorithm in the running state evaluation sub-chain. This step breaks through the limitations of traditional threshold detection by fusing the time-frequency domain characteristics of equipment running parameters and embedding the knowledge of historical maintenance data, so that the equipment implicit abnormalities can be identified by the feature matching model even when the equipment running evaluation score does not trigger an alarm. This evaluation mechanism based on multi-source data deep representation significantly improves the sensitivity and false alarm suppression capability of equipment state monitoring. Second, the process monitoring sub-chain constructs a full-process process abnormality evaluation network through the parameter constraint relationship between cross-process quality detection nodes and process node sub-chains. When the surface roughness Ra value is detected to be out of standard, the preset correlation influence degree in the hierarchical process index library is used to quickly locate to the pickling process equipment node through backward mapping connection, and the forward propagation simulation and reverse trace analysis of the bidirectional graph neural network are combined to realize the causal decoupling of process abnormalities and equipment implicit failures. This cross-layer correlation reasoning mechanism effectively solves the monitoring blind area problem of equipment normal but process abnormal. Third, the cascade design of loose coupling connection and collaborative causal process connection in the production process chain enables the process chain deviation caused by pickling concentration deviation to be gradually eliminated among multiple processes according to the influence weight through a distributed correction strategy, avoiding the secondary abnormal risk caused by traditional single-point parameter over-adjustment. The realization of this technical effect relies on the accurate modeling of process parameter transfer function and equipment control response characteristics. In addition, the multi-level signature verification mechanism enhanced by the blockchain and the parameter collaborative adjustment driven by the smart contract convert the process chain adjustment process into a digital certificate with legal effect, and through the feature sedimentation of the case library and the automatic generation of rules, a knowledge closed-loop evolution system is constructed, enabling the system to continuously adapt to new abnormal patterns. This technical fusion not only ensures the auditability and compliance of process adjustment, but also realizes the digital inheritance of experience knowledge through machine learning mechanism. Finally, the dynamic network topology and evidence weight allocation strategy of the Bayesian causal reasoning engine guide the correction strategy to preferentially trigger laser scanning re-inspection when detecting cold rolling size out-of-tolerance, and simultaneously adjust the pickling spray angle. This decision-making mechanism based on probability reasoning significantly improves the processing efficiency of complex abnormal scenarios.The synergy of the above technical means essentially builds an intelligent manufacturing closed-loop system with data perception, feature extraction, causal reasoning, dynamic correction and knowledge evolution. The implicit technical breakthrough is that through digital twin modeling of the process chain and autonomous learning of the causal correlation network, the traditional passive quality control based on the rule engine is upgraded to a cognitive intelligent system with predictive control capability, ensuring high-precision pipe production quality.

[0129] Embodiment 2

[0130] The application discloses an electronic device, comprising a memory and a processor, the memory stores a computer program, and the processor executes the computer program to realize an abnormality detection and processing method of an intelligent factory production process.

[0131] The application also discloses a computer readable storage medium, which stores computer instructions, and the computer instructions execute an abnormality detection and processing method of an intelligent factory production process when running.

[0132] The embodiments of the application are described above in combination with the drawings, but the application is not limited to the specific embodiments described above, and the specific embodiments described above are only illustrative but not restrictive. Those skilled in the art can make changes, modifications, replacements and variations to the above embodiments without departing from the purpose of the application and the scope protected by the claims under the inspiration of the application, and these are all within the protection of the application.

Claims

1. A method for anomaly detection and handling of a smart factory production process, characterized in that, The application relates to a production process chain monitoring method and device. The method comprises the following steps: Based on a preset production process chain, a double monitoring node chain and a hierarchical abnormality association evaluation node chain are constructed; The hierarchical abnormality association evaluation node chain comprises an operation state evaluation subchain and a whole-process process abnormality evaluation subchain; Real-time monitoring is performed on the production process chain based on the double monitoring node chain, and device operation state information and to-be-processed device production process information are obtained; Based on the device operation state information and the to-be-processed device production process information, a whole-process associated abnormality evaluation space is obtained by combining the hierarchical abnormality association evaluation node chain; Based on the whole-process associated abnormality evaluation space, a differential abnormality processing strategy is obtained by combining a historical abnormality strategy index information through a distributed algorithm, and the differential abnormality processing strategy is fed back to the double monitoring node chain in real time, so that real-time early warning and processing are performed on the monitored and positioned abnormality in the production process chain; The construction process of the production process chain comprises the following steps: Whole-process production process information of each to-be-processed device is obtained, and pre-trained production process analysis algorithms are combined to obtain continuous processing technology node sequences and processing sequence information corresponding to each to-be-processed device, required processing device sequences and processing device operation interaction information, mapping information and associated causal information of each processing technology and processing device; Based on the continuous processing technology node sequences and processing sequence information corresponding to each to-be-processed device, a directed loose coupling connection between each processing technology node corresponding to each to-be-processed device is constructed through a loose coupling algorithm; Based on the continuous processing technology node sequences and the directed loose coupling connection, a process node subchain corresponding to each to-be-processed device is obtained; Meanwhile, a device operation node sequence is constructed based on the required processing device sequence, and a bidirectional operation abnormality reasoning connection is constructed through a graph neural network by using the processing device operation interaction information; 2. The method of anomaly detection and handling of a smart factory production process according to claim 1, characterized in that, Based on the device operation node sequence and the bidirectional operation abnormality reasoning connection, a device operation node subchain is obtained. The production process chain comprises process node subchains and device operation node subchains; the double monitoring node chain comprises device operation monitoring subchains and process monitoring subchains; Each processing technology node in the process node subchain saves processing technology parameters of the corresponding node; each operation node in the device operation node subchain saves device control parameters of the corresponding processing device performing processing technology; The nodes in the device operation monitoring subchain and the device operation node subchain are in one-to-one correspondence; the nodes in the process node subchain and the process monitoring subchain are in one-to-one correspondence; 3. The method of anomaly detection and handling of a smart factory production process according to claim 2, characterized in that, Each monitoring node in the device operation monitoring subchain is in one-to-one correspondence with a running evaluation node in the running state evaluation subchain. Each process monitoring node in the process monitoring subchain is in one-to-one correspondence with each process evaluation node in the whole-process process abnormality evaluation subchain; Each running evaluation node in the running state evaluation subchain is connected to the process evaluation nodes in the whole-process process abnormality evaluation subchain in a one-to-many backward mapping mode. Each of the operation evaluation nodes is configured with a set of operation hierarchical evaluation indexes for evaluating the running state of the corresponding processing equipment; and each of the process evaluation nodes is configured with a set of hierarchical process evaluation indexes for evaluating the corresponding processing process; The set of operation hierarchical evaluation indexes and the set of hierarchical process evaluation indexes are respectively stored in preset hierarchical operation index libraries and hierarchical process index libraries, and the hierarchical process index libraries and the hierarchical operation index libraries are connected in a one-to-many directional connection manner through a preset directional connection.

4. The method of anomaly detection and handling of a smart factory production process according to claim 3, characterized in that, The construction process of the production process flow chain further includes: Suppose that the current process node sub-chain contains M continuous processing process nodes, the equipment running node sub-chain contains N equipment running nodes, and the process parameters corresponding to each processing process node are processed and run on only one equipment running node; Based on the assumption, the processing process node corresponding to the process parameter running on each equipment running node is set as the main processing process node of the equipment running node, and the remaining continuous processing process nodes after the main processing process node constitute the causal correlation process node sequence of the current equipment running node; Based on the main causal correlation information between the running state and the processing process result of each equipment running node and the corresponding main processing process node, a self-correlation algorithm is used to construct the main correlation degree between each equipment running node and the corresponding main processing process node, and the main connection relationship between the corresponding equipment running node and the main processing process node is constructed by using the obtained main correlation degree; Meanwhile, based on the cooperative running causal information between each equipment running node and the corresponding causal correlation process node sequence, a cross-correlation algorithm is used to obtain the cooperative causal correlation degree of each equipment running node for each processing process node in the corresponding causal correlation process node sequence.

5. The method of anomaly detection and handling of a smart factory production process according to claim 4, characterized in that, The construction process of the production process flow chain further includes: Based on the cooperative causal correlation degree of each equipment running node for each processing process node in the corresponding causal correlation process node sequence, a cooperative causal process connection between the equipment running node and the corresponding causal correlation process node sequence is constructed; Based on the main connection relationship and the cooperative causal process connection, the nodes corresponding to the process node sub-chain and the equipment running node sub-chain are connected to obtain the production process flow chain; Based on a Bayesian algorithm, a causal reasoning model is constructed and configured into the bidirectional running abnormal reasoning connection and the cooperative causal process connection to obtain the production process flow chain with causal reasoning.

6. The method of anomaly detection and handling of a smart factory production process according to claim 5, characterized in that, The construction process of the dual monitoring node chain includes: Based on the architecture mapping of the equipment running node sub-chain, an equipment running monitoring sub-chain with the same architecture is obtained, and a monitoring device is configured according to the running attribute corresponding to each equipment running node to obtain the equipment running parameter space corresponding to the equipment running node sub-chain; The equipment running parameter space includes a single running parameter sequence corresponding to each equipment running node and an interactive running parameter sequence corresponding between the equipment running nodes; Meanwhile, based on the architecture mapping of the process node sub-chain, a process monitoring sub-chain with the same architecture is obtained, and a corresponding monitoring device is configured according to the processing process information to obtain the processing process monitoring information space corresponding to the equipment running node sub-chain.

7. The method of anomaly detection and handling of a smart factory production process according to claim 6, characterized in that, The construction process of the hierarchical abnormal correlation evaluation node chain includes: The architecture mapping of the device operation node sub-chain obtains the running state evaluation sub-chain of the same architecture, and the architecture mapping of the process node sub-chain obtains the whole-process process abnormality evaluation sub-chain of the same architecture; The main connection relationship in the production process flow chain with causal reasoning is mapped to the corresponding nodes between the running state evaluation sub-chain and the whole-process process abnormality evaluation sub-chain, and a one-to-many backward mapping connection between the running state evaluation sub-chain and the whole-process process abnormality evaluation sub-chain is constructed, thereby obtaining a hierarchical abnormality correlation evaluation node chain.

8. The method of anomaly detection and handling of a smart factory production process according to claim 7, characterized in that, The difference abnormality processing strategy acquisition process includes: Based on the dual monitoring node chain, the device operation parameter space and the processing process monitoring information space are monitored and acquired in real time, and are synchronously mapped to the corresponding running state evaluation sub-chain and whole-process process abnormality evaluation sub-chain in the hierarchical abnormality correlation evaluation node chain; Based on the running hierarchical evaluation index set configured in each running evaluation node in the running state evaluation sub-chain and the whole-process process abnormality evaluation sub-chain and the hierarchical process evaluation index set configured in each process evaluation node, the running state and the processing process result of each running evaluation node and the corresponding main processing process node are synchronously evaluated, thereby obtaining a running state evaluation result sequence and a processing process evaluation result sequence; The running state abnormality evaluation threshold sequence and the processing process abnormality evaluation threshold sequence are set, and based on the running state evaluation result sequence and the processing process evaluation result sequence, the running state abnormality evaluation threshold sequence and the processing process abnormality evaluation threshold sequence are combined to synchronously discriminate the device operation node sub-chain and the process node sub-chain, thereby obtaining a whole-process correlation abnormality evaluation space.

Images