Industrial equipment maintenance method and platform based on multi-source heterogeneous data

Abstract

The present invention relates to the field of industrial big data processing, and specifically to an industrial equipment maintenance method and platform based on multi-source heterogeneous data, the method comprising: acquiring multi-source heterogeneous data, including equipment operation and maintenance data, monitoring data, factory operation data and monitoring image data; generating equipment degradation health indicators based on the operation and maintenance data, and quantitatively indicating the health status of the equipment; performing feature extraction on the factory operation data and generating feature changes; constructing a convolutional neural network model, inputting monitoring images into the model, and outputting violation behavior images; performing maintenance operations on factory equipment according to the equipment degradation health indicators, factory operation data feature changes and violation behavior images; the method comprehensively utilizes multi-source data to realize intelligent monitoring and maintenance of equipment from three dimensions: equipment status, factory operation and personnel behavior, so as to improve equipment reliability, extend service life and ensure safe and efficient operation of the factory.

Description

Technical Field

[0001] The present invention relates to the field of industrial big data processing, and specifically to an industrial equipment maintenance method and platform based on multi-source heterogeneous data. Background Art

[0002] Industrial Internet and industrial big data provide new opportunities for the digital and intelligent transformation of factories. The massive data generated during the factory production process contains valuable knowledge and insights. Mining this data can optimize production processes, predict equipment failures, improve product quality and reduce operating costs. However, factory data is of various types, sources and formats, which brings huge challenges to data collection, processing and analysis.

[0003] Traditional data processing methods mainly target structured data, such as relational databases and data warehouses, but are difficult to effectively process unstructured data in factory environments, such as equipment logs, sensor readings, and video surveillance data. In addition, heterogeneous data from different sources differ in spatiotemporal granularity and semantic representation, making it difficult to directly fuse and correlate data.

[0004] In addition, the management and application of factory knowledge also face challenges. The knowledge of factory equipment, processes, materials and other fields is scattered in various documents and systems, lacking effective representation and organization methods. Knowledge acquisition is highly dependent on domain experts and lacks automated means.

[0005] Therefore, there is an urgent need for an intelligent processing method for multi-source heterogeneous industrial data, which is of great significance for promoting the development and application of industrial big data technology and realizing intelligent optimization and decision support of industrial production. Summary of the invention

[0006] The purpose of the present invention is to address the deficiencies of the prior art and to propose an industrial equipment maintenance method and platform based on multi-source heterogeneous data.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] In an embodiment of the present application, an industrial equipment maintenance method based on multi-source heterogeneous data is provided, including:

[0009] Acquire multi-source heterogeneous data, wherein the multi-source heterogeneous data includes industrial equipment operation and maintenance data, industrial equipment monitoring data, factory operation data, and monitoring image data; generate industrial equipment degradation health indicators based on the industrial equipment operation and maintenance data, wherein the equipment degradation health indicators include equipment efficiency decline rate and failure rate increase rate, and the equipment degradation health indicators are used to quantitatively indicate the health status of the equipment;

[0010] Extracting features from the acquired plant operation data to generate feature changes of the plant operation data;

[0011] Constructing a convolutional neural network model, inputting the monitoring image data into the convolutional neural network model, and outputting an image of the violation behavior;

[0012] Maintenance operations are performed on the industrial equipment of the factory according to the degradation health index of the industrial equipment, the characteristic changes of the factory operation data, and the violation behavior image.

[0013] In a preferred embodiment, the factory operation data is collected, and the production efficiency data is collected in real time. , Resource consumption data , resource supplement data , integrate the running data into a time series dataset , , ; is the sampling time step;

[0014] Through the video stream access of the surveillance camera, the factory monitoring video frame is periodically extracted; let the image collected by the j-th camera at time t be , representing the image data stream as a time series , where S is the number of cameras;

[0015] The data collected by the sensor is preprocessed, and the preprocessing includes missing value filling, noise removal and time alignment.

[0016] In a preferred embodiment, the feature extraction includes time domain statistical feature calculation, data change trend calculation and data correlation calculation.

[0017] In a preferred embodiment, the industrial equipment operation and maintenance data includes industrial equipment maintenance records and industrial equipment failure history; the industrial equipment degradation health index also includes the number of equipment maintenance times, and the industrial equipment degradation health index is further used to quantify the remaining life of the equipment;

[0018] The remaining life of the equipment is defined as the time interval from the current moment to the equipment failure, denoted as ,in is the equipment failure time, is the current time;

[0019] The industrial equipment maintenance method further comprises:

[0020] Construct a training set sample of the long short-term memory (LSTM) network: for each device, extract sensor monitoring data and operation and maintenance data; the sensor monitoring data includes the vibration data, temperature data and pressure data of the device; divide the operation time of the device into multiple time windows of fixed length, each window contains time steps, and set the sliding step size of the window to For the time window, extract the sensor monitoring data and event data from it, and form a sample ,in For the The feature vector of each time step;

[0021] Calculate the The end time of the time window When the equipment fails The time interval, where , as the remaining lifetime label of the sample ;

[0022] Traverse all time windows to get the training set of the device ; where N is the number of samples.

[0023] In a preferred embodiment, the method further includes: performing uncertainty quantification and confidence assessment on the remaining life prediction of the factory equipment, wherein the uncertainty quantification includes introducing the Monte Carlo neuron random inactivation method when quantifying the remaining life of the equipment based on the long short-term memory (LSTM) network, and performing the confidence assessment on the training set. Repeat for the kth sample in Forward inference, each time randomly inactivating some neurons, we get The remaining life expectancy is The mean of the remaining life prediction values and variance for:

[0024]

[0025] in, For the The remaining life prediction value obtained by forward extrapolation;

[0026] The confidence assessment includes designing a confidence scoring function, and performing a confidence assessment on the prediction result of the remaining life of the plant equipment based on the mean, variance and historical prediction error of the predicted value. The confidence scoring function is:

[0027]

[0028] in, is the confidence evaluation score, and the confidence score range is [0,1]; EWMA is the exponentially weighted moving average of historical forecast errors, is the predicted value of the remaining life of the equipment, is the confidence factor.

[0029] In a preferred embodiment, the constructing of a convolutional neural network model, inputting the monitoring image data into the convolutional neural network model, and outputting a violation behavior image comprises:

[0030] Construct convolutional neural network (CNN) and recurrent neural network (RNN) to identify and extract images of violations in factory surveillance videos;

[0031] Based on the training set with images of illegal behaviors, the convolutional neural network (CNN) is trained to identify images of illegal behaviors in factory monitoring videos; the training process includes: manually dividing the normal behavior image labels and illegal behavior image labels in the training set, training the convolutional neural network (CNN) to output the proportion of the two labels in the training set, constructing a training loss function, and completing the training when the loss function converges;

[0032] Based on the training set constructed based on the violation images, the recurrent neural network RNN ​​is trained to identify the temporal characteristics of the violation; multiple violation images are constructed into a complete violation action through the recurrent neural network RNN.

[0033] In a preferred embodiment, the industrial equipment maintenance method further includes: building a large language model to generate textual descriptions of characteristic changes in factory operation data, violation behavior images, and equipment degradation health indicators, and generating a factory text report based on the textual descriptions;

[0034] Collect text data on factory equipment operation, maintenance, and production management to build an industry domain corpus;

[0035] Use the pre-trained language model BERT and train it on the domain corpus to learn industry-specific knowledge and language habits;

[0036] The output text of the feature change detection, violation detection and equipment health indicator prediction models is used as the input of the report generation model to design a framework for multi-document summary generation.

[0037] A hierarchical attention mechanism is introduced into the framework to automatically learn how to aggregate information at different text granularities, including words, sentences, and paragraphs.

[0038] The Transformer self-attention mechanism model is used as the decoder to generate the factory operation report.

[0039] In a preferred embodiment, the industrial equipment maintenance method further comprises:

[0040] Define entity types in the factory domain, use the conditional random field (CRF) sequence labeling model to identify entities in the text; use domain dictionaries and rule templates to improve the accuracy of entity recognition;

[0041] Define the relationship types in the factory field, use the PCNN relationship extraction model based on neural network to extract the relationship between entities from the text; deduplicate and merge the identified entities and relationships to eliminate redundant information; use external knowledge base to disambiguate and complete entities and relationships; embed entities and relationships into a low-dimensional semantic space through the knowledge representation learning method TransE; convert the extracted entities and relationships into nodes and edges of the knowledge graph; use a graph database to store and manage the knowledge graph.

[0042] The present application further provides an industrial equipment maintenance platform based on multi-source heterogeneous data, specifically including: a data acquisition module, an equipment health quantification module, a data feature extraction module, and a violation detection module;

[0043] The data acquisition module is used to acquire multi-source heterogeneous data, wherein the multi-source heterogeneous data includes industrial equipment operation and maintenance data, industrial equipment monitoring data, factory operation data and monitoring image data;

[0044] The equipment health quantification module generates an industrial equipment degradation health index based on the industrial equipment operation and maintenance data, wherein the equipment degradation health index includes an equipment efficiency decline rate and a failure rate increase rate, and the equipment degradation health index is used to quantitatively indicate the equipment health status;

[0045] The data feature extraction module is used to extract features from the acquired plant operation data and generate feature changes of the plant operation data;

[0046] The violation detection module is used to construct a convolutional neural network model, input the monitoring image data into the convolutional neural network model, and output a violation image.

[0047] An electronic device comprises: a processor and a memory, wherein the memory stores a computer program that can be called by the processor;

[0048] The processor executes the industrial equipment maintenance method based on multi-source heterogeneous data by calling the computer program stored in the memory.

[0049] A computer-readable storage medium includes: instructions stored therein, and when the instructions are executed on a computer, the computer is enabled to execute the industrial equipment maintenance method based on multi-source heterogeneous data.

[0050] Beneficial effects of the present invention: The present invention obtains multi-source heterogeneous data to fully understand the operating conditions of industrial equipment and factories, providing a rich data basis for subsequent analysis.

[0051] The present invention generates equipment degradation health indicators, quantitatively indicates the health status of equipment, and provides a basis for predicting equipment failures and formulating maintenance strategies.

[0052] The present invention performs feature extraction on factory operation data, obtains data features and change trends in the factory data, helps to discover abnormal situations during factory operation, and take timely measures to ensure production efficiency and quality.

[0053] The present invention constructs a convolutional neural network model, inputs monitoring image data into the model, and outputs images of illegal behavior, which can automatically identify illegal operations in the industrial production process, improve the level of safety management, and reduce the risk of accidents caused by human factors.

[0054] The present invention comprehensively considers factors such as equipment health status, factory operation characteristics and violations, performs targeted maintenance on industrial equipment, improves equipment reliability, reduces downtime, and ensures factory production efficiency and safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a flow chart of the industrial equipment maintenance method based on multi-source heterogeneous data of the present invention;

[0056] Figure 2 This is a structural diagram of the industrial equipment maintenance platform based on multi-source heterogeneous data of the present invention;

[0057] Figure 3 It is a schematic diagram of the structure of an electronic device provided by the present invention;

[0058] Figure 4 It is a schematic diagram of the structure of the computer-readable storage medium provided by the present invention. DETAILED DESCRIPTION

[0059] In order to better understand the present application, a more detailed description will be made of various aspects of the present application with reference to the accompanying drawings. It should be understood that these detailed descriptions are only descriptions of exemplary embodiments of the present application, and are not intended to limit the scope of the present application in any way. Throughout the specification, the same reference numerals refer to the same elements. The expression "and / or" includes any and all combinations of one or more of the associated listed items.

[0060] In the accompanying drawings, the size, dimensions and shapes of the elements have been slightly adjusted for ease of illustration. The drawings are for illustration only and are not strictly drawn to scale. As used herein, the terms "substantially", "approximately" and similar terms are used as terms of approximation, not as terms of degree, and are intended to illustrate the inherent deviations in measurements or calculations that will be recognized by those of ordinary skill in the art. In addition, in this application, the order in which the steps are described does not necessarily represent the order in which these processes occur in actual operation, unless otherwise specified or can be derived from the context.

[0061] It should also be understood that expressions such as "include", "including", "have", "contain" and / or "comprising" are open rather than closed expressions in this specification, which indicate the presence of the stated features, elements and / or components, but do not exclude the presence of one or more other features, elements, components and / or combinations thereof. In addition, when expressions such as "at least one of..." appear after a list of listed features, they modify the entire list of features rather than just the individual elements in the list. In addition, when describing embodiments of the present application, "may" is used to mean "one or more embodiments of the present application". And, the term "exemplary" is intended to refer to an example or illustration.

[0062] Unless otherwise defined, all words (including engineering terms and scientific and technological terms) used in this article have the same meaning as those commonly understood by ordinary technicians in the field to which this application belongs. It should also be understood that unless there is a clear explanation in this application, words defined in commonly used dictionaries should be interpreted as having the same meaning as their meaning in the context of the relevant technology, and should not be interpreted in an idealized or overly formal sense.

[0063] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0064] Example 1

[0065] Reference Figure 1 ,The first embodiment of the present invention provides an industrial equipment ,maintenance method based on multi-source heterogeneous data.

[0066] S1: Acquire multi-source heterogeneous data, where the multi-source heterogeneous data includes industrial equipment operation and maintenance data, industrial equipment monitoring data, factory operation data, and monitoring image data.

[0067] Specifically, multi-source heterogeneous data has different output interfaces, different output types, and different data feature dimensions. This application does not elaborate on the definition of multi-source heterogeneous data. It needs to be understood that in the industrial production process, multi-source heterogeneous data can be determined based on the type of factory equipment or the type of process scenario. This application does not limit or exhaustively list this.

[0068] Specifically, collect factory operation data and real-time production efficiency data , Resource consumption data , resource supplement data , integrate the running data into a time series dataset , , ; is the number of sampling time steps.

[0069] Through the video stream access of the surveillance camera, the factory monitoring video frame is periodically extracted; let the image collected by the j-th camera at time t be , representing the image data stream as a time series , where S is the number of cameras.

[0070] The data collected by the sensor is preprocessed, and the preprocessing includes missing value filling, noise removal and time alignment.

[0071] The missing value filling method uses linear interpolation to estimate the missing value and minimize the estimation error. ,in is the set of time steps with missing values, is the set of time series of running data after filling missing values.

[0072] The noise removal is to use a moving average filter to smooth the running data, that is, , where K is the filter window radius, k is the index variable of the summation symbol, representing the value from -K to K.

[0073] The time alignment is to resample the running data with inconsistent sampling frequencies so as to align them to the same time step length.

[0074] The image data preprocessing includes image enhancement and image size scaling. The image enhancement is to adjust the brightness and contrast of the image captured by the monitoring camera to the same level.

[0075] The image size scaling is to uniformly scale the images of different resolutions collected by the surveillance camera to a fixed size.

[0076] The collected plant operation data features are extracted, and the feature extraction includes time domain statistical feature calculation, data change trend calculation and data correlation calculation.

[0077] The time domain statistical feature calculation includes: mean , standard deviation ; Wherein, i is a parameter, representing the index of the type of factory operation data.

[0078] The data change trend calculation includes: Do linear regression and fit the slope and the intercept ,Right now ; Slope Represents the rate of change of the operating data of the i-th type of plant at time t.

[0079] The data correlation calculation includes: calculating production efficiency data , Resource consumption data , resource supplement data The Pearson correlation coefficient between .

[0080] Production efficiency data Resource consumption data Pearson correlation coefficient between The calculation is as follows:

[0081]

[0082] The Pearson correlation coefficient is calculated in the same way. and ; represents the mean of production efficiency data, It represents the mean of resource consumption data. The Pearson correlation coefficient takes values ​​between [-1, 1], indicating the linear correlation strength between the two indicators. The closer the Pearson correlation coefficient is to 1, the stronger the correlation between the two data, and the closer it is to -1, the weaker the correlation.

[0083] By acquiring multi-source heterogeneous data, including industrial equipment operation and maintenance data, monitoring data, factory operation data, and monitoring image data, and preprocessing and feature extraction of the data, we can fully understand the status of industrial equipment and factory operation, eliminate missing values, noise interference and other problems in the data, and extract representative and discriminative features. This provides a high-quality data foundation for subsequent analysis tasks such as equipment health status assessment, anomaly detection, and fault prediction, thereby improving the accuracy and reliability of the analysis results.

[0084] S2: Based on the industrial equipment operation and maintenance data, generate an industrial equipment degradation health indicator, wherein the equipment degradation health indicator includes an equipment efficiency decrease rate and a failure rate increase rate, and the equipment degradation health indicator is used to quantitatively indicate the equipment health status.

[0085] Collecting operation and maintenance data and sensor monitoring data of factory equipment, wherein the operation and maintenance data includes maintenance records and failure history; constructing equipment degradation health indicators, wherein the equipment degradation health indicators are the number of equipment maintenance times and quantifying the remaining life of the equipment.

[0086] The remaining life of the equipment is defined as the time interval from the current moment to the equipment failure, denoted as ,in is the equipment failure time, is the current time.

[0087] Construct a training set sample of the long short-term memory (LSTM) network: for each device, extract sensor monitoring data and operation and maintenance data; the sensor monitoring data includes the vibration data, temperature data and pressure data of the device; divide the operation time of the device into multiple time windows of fixed length, each window contains time steps, and set the sliding step size of the window to For the time window, extract the sensor monitoring data and event data from it, and form a sample ,in For the The feature vector of each time step.

[0088] Calculate the The end time of the time window When the equipment fails The time interval, where , as the remaining lifetime label of the sample .

[0089] Traverse all time windows to get the training set of the device ; where N is the number of samples.

[0090] Long short-term memory (LSTM) network models the complex nonlinear mapping relationship between the equipment degradation process and the remaining life. The training steps of the LSTM network are as follows:

[0091] The operation and maintenance data of the equipment is divided into multiple degradation cycles, each cycle ends with equipment failure, the data points before the failure are used as input, and the time interval to the failure is used as the label.

[0092] Optimize model parameters using stochastic gradient descent , minimize the loss function ; The parameter update formula is ,in is the learning rate, represents the gradient of the loss function with respect to the model parameters θ.

[0093] The Bayesian optimization algorithm is used to search for the optimal model hyperparameters, including the optimal number of LSTM layers, number of hidden units, and learning rate.

[0094] The Bayesian optimization algorithm is used to search for the optimal LSTM network hyperparameters and confidence factors. The goal is to maximize the expectation of the confidence score function. The acquisition function of Bayesian optimization is:

[0095]

[0096] in, is the posterior mean of the confidence score function obtained by Gaussian process regression, is the posterior variance, is a trade-off factor.

[0097] The uncertainty quantification and confidence assessment of the remaining life prediction of factory equipment are carried out. The uncertainty quantification includes the introduction of the Monte Carlo neuron random inactivation method when quantifying the remaining life of the equipment based on the long short-term memory (LSTM) network, and the training set Repeat for the kth sample in Forward inference, each time randomly inactivating some neurons, we get The remaining life expectancy is The mean of the remaining life prediction values and variance for:

[0098]

[0099] in, For the The remaining life prediction value obtained by forward extrapolation.

[0100] The confidence assessment includes designing a confidence scoring function, and performing a confidence assessment on the prediction result of the remaining life of the plant equipment based on the mean, variance and historical prediction error of the predicted value. The confidence scoring function is:

[0101]

[0102] in, is the confidence evaluation score, and the confidence score range is [0,1]; EWMA is the exponentially weighted moving average of historical forecast errors, is the predicted value of the remaining life of the equipment, is the confidence factor.

[0103] Based on the operation and maintenance data of industrial equipment, the equipment degradation health indicators are generated, including the equipment efficiency decline rate and the failure rate increase rate. The long short-term memory (LSTM) network is used to model the complex nonlinear mapping relationship between the equipment degradation process and the remaining life. Combined with the Bayesian optimization algorithm to search for the optimal model hyperparameters, the health status of the equipment can be accurately quantified, the remaining life of the equipment can be predicted, and the uncertainty of the prediction results can be quantified and the confidence level can be evaluated. This helps factory managers to grasp the degradation trend of equipment in a timely manner, arrange equipment maintenance and update plans reasonably, avoid production shutdowns and economic losses caused by sudden equipment failures, and improve equipment reliability and production efficiency.

[0104] S3: Extract features from the acquired plant operation data to generate feature changes of the plant operation data.

[0105] Convolutional neural network (CNN) and recurrent neural network (RNN) are constructed to identify and extract images of violations in factory surveillance videos.

[0106] Based on the training set with violation images, the convolutional neural network (CNN) is trained to identify violation images in factory monitoring videos. The training process includes: manually dividing the normal behavior image labels and violation image labels in the training set, training the convolutional neural network (CNN) to output the proportion of the two labels in the training set, constructing a training loss function, and completing the training when the loss function converges.

[0107] Based on the training set constructed based on the violation images, the recurrent neural network RNN ​​is trained to identify the temporal characteristics of the violation; multiple violation images are constructed into a complete violation action through the recurrent neural network RNN.

[0108] The model structure of the constructed convolutional neural network CNN is: multi-layer convolution and pooling operations are used to extract image semantic features; the output of the vth layer convolution is ,in is the convolution kernel parameter, is the bias term, * is the convolution operation, is an activation function (such as ReLU); the pooling operation downsamples the convolution output to reduce the size of the feature map; and the convolutional neural network CNN is used to extract the violation images in the factory monitoring video.

[0109] The RNN variant of the GRU model is used to model the feature sequence of the violation image; the final moment is output The hidden state of Contains the temporal characteristics of the entire violation sequence.

[0110] Using cross entropy loss function As the loss function of the convolutional neural network CNN; is the true label of the i-th sample belonging to category c, is the probability predicted by the model.

[0111] Use the back-propagation algorithm to update the parameters of CNN and RNN, minimize the loss function as the training goal, and stop training when the loss function converges.

[0112] By building a convolutional neural network (CNN) and a recurrent neural network (RNN), we can identify and extract images of illegal behaviors in factory surveillance videos, automatically discover workers' illegal operations, such as not wearing a helmet, not operating equipment according to specifications, etc., and build multiple illegal behavior images into a complete illegal behavior action through time series feature analysis. This helps factory managers monitor safety hazards at the production site in real time, take corrective measures in a timely manner, improve workers' safety awareness and standardized operation level, reduce the risk of safety accidents, and protect workers' lives and corporate property safety.

[0113] S4: Construct a convolutional neural network model, input the monitoring image data into the convolutional neural network model, and output the violation image.

[0114] Build a large language model to generate textual descriptions of characteristic changes in factory operation data, violation behavior images, and equipment degradation health indicators, and generate factory text reports based on the textual descriptions.

[0115] Collect text data related to factory equipment operation, maintenance, and production management to build an industry corpus.

[0116] The pre-trained language model BERT is used and trained on the domain corpus to learn industry-specific knowledge and language habits.

[0117] The output text of feature change detection, violation detection and equipment health indicator prediction models is used as the input of the report generation model, and a framework for multi-document summary generation is designed.

[0118] A hierarchical attention mechanism is introduced into the framework to automatically learn information aggregation methods of different text granularities, including words, sentences, and paragraphs.

[0119] The Transformer self-attention mechanism model is used as the decoder to generate the factory operation report.

[0120] By building a large language model, we can automatically generate text descriptions of characteristic changes, violations, and equipment health status of factory operation data, and intelligently generate factory operation reports based on these descriptions. This can greatly improve the efficiency and accuracy of report writing, provide factory managers with more timely, comprehensive, and reliable decision-making basis, and help factories achieve digital transformation and intelligent operations.

[0121] Template guidance and knowledge enhancement technology are integrated into the generation process to ensure the professionalism and accuracy of the report content.

[0122] Design a feedback mechanism to allow users to evaluate and modify the generated reports, and use it to iteratively optimize the report generation model.

[0123] By building a large language model, we can automatically generate text descriptions of characteristic changes, violations, and equipment health status of factory operation data, and intelligently generate factory operation reports based on these descriptions, which can greatly improve the efficiency and accuracy of report writing. The traditional manual report writing method is time-consuming and labor-intensive, and is subject to the professional level and experience of the writers, and the quality of reports varies. Intelligent report generation technology can quickly and accurately mine key information and insights from massive factory data, generate professional, comprehensive, and objective analysis reports, and provide factory managers with more timely and reliable decision-making basis, helping factories achieve data-driven scientific management and intelligent operations. At the same time, by introducing a user feedback mechanism, the report generation model can be continuously iterated and optimized to better adapt to the actual needs and business scenarios of the factory.

[0124] S5: Perform maintenance operations on the industrial equipment of the factory according to the degradation health index of the industrial equipment, the characteristic changes of the factory operation data, and the violation behavior image.

[0125] Define entity types in the factory field, such as equipment, parts, materials, personnel, processes, etc.; use the conditional random field (CRF) sequence labeling model to identify entities in the text; use domain dictionaries, rule templates and other methods to improve the accuracy of entity recognition.

[0126] Define the relationship types in the factory field, such as equipment-parts, materials-processes, personnel-operations, etc.; use the PCNN relationship extraction model based on neural network to extract the relationship between entities from the text.

[0127] De-duplicate and merge the identified entities and relationships to eliminate redundant information; use external knowledge bases, such as equipment manuals, process flow charts, etc., to disambiguate and complete entities and relationships.

[0128] Through the knowledge representation learning method TransE, entities and relations are embedded into a low-dimensional semantic space.

[0129] Convert the extracted entities and relationships into nodes and edges of the knowledge graph; use graph databases, such as Neo4j, JanusGraph, etc., to store and manage the knowledge graph.

[0130] Design the query interface and visualization interface of the knowledge graph to facilitate users to explore and analyze factory industry data.

[0131] Use knowledge graphs for applications such as equipment fault diagnosis, production process optimization, and quality problem tracing; combine knowledge graphs with machine learning models to achieve knowledge-driven prediction and decision support.

[0132] By identifying and extracting entity relationships from factory text reports and building a knowledge graph, we can deeply mine the associated information and hidden insights contained in the factory's industrial data, and provide intelligent data support and decision-making basis for the factory's equipment management, production optimization, quality control, etc., helping factories achieve digital transformation and smart manufacturing upgrades, and improve operational efficiency and product competitiveness.

[0133] Example 2

[0134] Reference Figure 2 ,The second embodiment of the present invention provides an industrial equipment ,maintenance platform based on multi-source heterogeneous data.

[0135] The platform includes: a data acquisition module, an equipment health quantification module, a data feature extraction module and a violation detection module.

[0136] The data acquisition module is used to acquire multi-source heterogeneous data, where the multi-source heterogeneous data includes industrial equipment operation and maintenance data, industrial equipment monitoring data, factory operation data, and monitoring image data.

[0137] The equipment health quantification module generates industrial equipment degradation health indicators based on the industrial equipment operation and maintenance data. The equipment degradation health indicators include equipment efficiency decline rate and failure rate increase rate, and the equipment degradation health indicators are used to quantitatively indicate the health status of the equipment.

[0138] The data feature extraction module is used to extract features from the acquired plant operation data and generate feature changes of the plant operation data.

[0139] The violation detection module is used to construct a convolutional neural network model, input the monitoring image data into the convolutional neural network model, and output a violation image.

[0140] Example 3

[0141] Figure 3 FIG. 1 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. Figure 3 As shown, according to another aspect of the present invention, an electronic device 500 is also provided. The electronic device 500 may include one or more processors and one or more memories. The memories store computer readable codes, and when the computer readable codes are executed by one or more processors, the industrial equipment maintenance method based on multi-source heterogeneous data as described above may be executed.

[0142] The method or system according to the embodiment of the present invention can also be used by Figure 3 The architecture of the electronic device shown is implemented.

[0143] like Figure 3As shown, the electronic device 500 may include a bus 501, one or more CPUs 502, a read-only memory (ROM) 503, a random access memory (RAM) 504, a communication port 505 connected to a network, an input / output component 506, a hard disk 507, and the like.

[0144] The storage device in the electronic device 500, such as ROM503 or hard disk 507, can store the industrial equipment maintenance method based on multi-source heterogeneous data provided by the present invention, including: obtaining multi-source heterogeneous data, the multi-source heterogeneous data including industrial equipment operation and maintenance data, industrial equipment monitoring data, factory operation data and monitoring image data; based on the industrial equipment operation and maintenance data, generating industrial equipment degradation health indicators, the equipment degradation health indicators including equipment efficiency decline rate and failure rate increase rate, and the equipment degradation health indicators are used to quantitatively indicate the health status of the equipment; extracting features from the acquired factory operation data to generate feature changes in the factory operation data; constructing a convolutional neural network model, inputting the monitoring image data into the convolutional neural network model, and outputting a violation behavior image; performing maintenance operations on the industrial equipment of the factory according to the industrial equipment degradation health indicators, the feature changes in the factory operation data and the violation behavior image. Furthermore, the electronic device 500 may also include a user interface 508. Of course, Figure 3 The architecture shown is only exemplary and can be omitted according to actual needs when implementing different devices. Figure 3 One or more components of an electronic device are shown.

[0145] Example 4

[0146] Figure 4 It is a schematic diagram of the structure of a computer-readable storage medium provided by an embodiment of the present invention.

[0147] like Figure 4 As shown, a computer readable storage medium 600 according to one embodiment of the present invention.

[0148] Computer readable storage medium 600 has computer readable instructions stored thereon.

[0149] When the computer-readable instructions are executed by a processor, the industrial equipment maintenance method based on multi-source heterogeneous data according to the embodiments of the present invention described with reference to the above drawings can be executed.

[0150] The storage medium 600 includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache), etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. In addition, according to an embodiment of the present invention, the process described above with reference to the flowchart may be implemented as a computer software program.

[0151] For example, the present invention provides a non-transitory machine-readable storage medium, which stores machine-readable instructions, and the machine-readable instructions can be executed by a processor to execute instructions corresponding to the method steps provided by the present invention, for example: obtaining multi-source heterogeneous data, the multi-source heterogeneous data including industrial equipment operation and maintenance data, industrial equipment monitoring data, factory operation data and monitoring image data; based on the industrial equipment operation and maintenance data, generating industrial equipment degradation health indicators, the equipment degradation health indicators including equipment efficiency decline rate and failure rate increase rate, and the equipment degradation health indicators are used to quantitatively indicate the health status of the equipment; extracting features from the acquired factory operation data to generate feature changes in the factory operation data; constructing a convolutional neural network model, inputting the monitoring image data into the convolutional neural network model, and outputting a violation behavior image; performing maintenance operations on the industrial equipment of the factory according to the industrial equipment degradation health indicators, the feature changes in the factory operation data and the violation behavior image. When the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present invention are executed. The method, apparatus and device of the present invention may be implemented in many ways. For example, the methods, apparatuses, and devices of the present invention may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware.

[0152] The above sequence for the steps of the method is for illustration only, and the steps of the method of the present invention are not limited to the sequence specifically described above unless otherwise specifically stated.

[0153] In addition, in some embodiments, the present invention can also be implemented as a program recorded in a recording medium, which includes machine-readable instructions for implementing the method according to the present invention. Therefore, the present invention also covers a recording medium storing a program for executing the method according to the present invention.

[0154] In addition, the parts of the above technical solutions provided in the embodiments of the present invention that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive redundancy.

[0155] The specific implementation modes as described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation mode of the present invention and is not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An industrial equipment maintenance method based on multi-source heterogeneous data, characterized in that: include: Acquire multi-source heterogeneous data, wherein the multi-source heterogeneous data includes industrial equipment operation and maintenance data, industrial equipment monitoring data, factory operation data, and monitoring image data; Based on the industrial equipment operation and maintenance data, generate an industrial equipment degradation health index, wherein the equipment degradation health index includes an equipment efficiency decline rate and a failure rate increase rate, and the equipment degradation health index is used to quantitatively indicate the health status of the equipment; Extracting features from the acquired plant operation data to generate feature changes of the plant operation data; Constructing a convolutional neural network model, inputting the monitoring image data into the convolutional neural network model, and outputting an image of the violation behavior; Perform maintenance operations on the industrial equipment of the factory according to the degradation health index of the industrial equipment, the characteristic changes of the factory operation data, and the violation behavior image; Collect factory operation data and real-time production efficiency data , Resource consumption data , resource supplement data , integrate the running data into a time series dataset , , ; is the sampling time step; Through the video stream access of the surveillance camera, the factory monitoring video frame is periodically extracted; let the image collected by the j-th camera at time t be , representing the image data stream as a time series , , where S is the number of cameras; Preprocessing the data collected by the sensor, wherein the preprocessing includes missing value filling, noise removal and time alignment; The feature extraction includes time domain statistical feature calculation, data change trend calculation and data correlation calculation; The time domain statistical feature calculation includes: mean , standard deviation ; Where i is a parameter, indicating the index of the type of factory operation data; The data change trend calculation includes: Do linear regression and fit the slope and the intercept ,Right now ; Slope represents the rate of change of the operating data of the i-th type of plant at time t; The data correlation calculation includes: calculating production efficiency data , Resource consumption data , resource supplement data Pearson correlation coefficient between ; Production efficiency data Resource consumption data Pearson correlation coefficient between The calculation is as follows: ； The Pearson correlation coefficient is calculated in the same way and ; represents the mean of production efficiency data, Represents the mean of resource consumption data; The industrial equipment operation and maintenance data includes industrial equipment maintenance records and industrial equipment failure history; the industrial equipment degradation health index also includes the number of equipment maintenance times, and the industrial equipment degradation health index is further used to quantify the remaining life of the equipment; The remaining life of the equipment is defined as the time interval from the current moment to the equipment failure, denoted as ,in is the equipment failure time, is the current time; The industrial equipment maintenance method also includes: constructing a training set sample of a long short-term memory (LSTM) network: for each device, extracting sensor monitoring data and operation and maintenance data; the sensor monitoring data includes vibration data, temperature data and pressure data of the device; dividing the operation time of the device into multiple time windows of fixed length, each window contains time steps, and set the sliding step size of the window to For the time window, extract the sensor monitoring data and event data from it, and form a sample ,in For the The feature vector of each time step; Calculate the The end time of the time window When the equipment fails The time interval, where , as the remaining life label of the sample ; Traverse all time windows to get the training set of the device ; Where N is the number of samples; The method also includes: performing uncertainty quantification and confidence assessment on the remaining life prediction of the factory equipment, wherein the uncertainty quantification includes introducing a Monte Carlo neuron random inactivation method when quantifying the remaining life of the equipment based on a long short-term memory (LSTM) network, and performing a random inactivation on the training set. Repeat for the kth sample in Forward inference, each time randomly inactivating some neurons, we get The remaining life expectancy is The mean of the remaining life prediction values and variance for: ； in, For the The remaining life prediction value obtained by forward extrapolation; The confidence assessment includes designing a confidence scoring function, and performing a confidence assessment on the prediction result of the remaining life of the plant equipment based on the mean, variance and historical prediction error of the predicted value. The confidence scoring function is: ； in, is the confidence evaluation score, and the confidence score range is [0,1]; EWMA is the exponentially weighted moving average of historical forecast errors, is the predicted value of the remaining life of the equipment, is the confidence factor.

2. The industrial equipment maintenance method based on multi-source heterogeneous data according to claim 1 is characterized in that: The constructing of the convolutional neural network model, inputting the monitoring image data into the convolutional neural network model, and outputting the violation behavior image includes: constructing a convolutional neural network CNN and a recurrent neural network RNN ​​for identifying and extracting the violation behavior image in the factory monitoring video; Based on the training set with images of illegal behaviors, the convolutional neural network (CNN) is trained to identify images of illegal behaviors in factory monitoring videos; the training process includes: manually dividing the normal behavior image labels and illegal behavior image labels in the training set, training the convolutional neural network (CNN) to output the proportion of the two labels in the training set, constructing a training loss function, and completing the training when the loss function converges; Based on the training set constructed based on the violation images, the recurrent neural network RNN ​​is trained to identify the temporal characteristics of the violation; multiple violation images are constructed into a complete violation action through the recurrent neural network RNN.

3. The industrial equipment maintenance method based on multi-source heterogeneous data according to claim 2 is characterized in that: The industrial equipment maintenance method further includes: building a large language model to generate textual descriptions of characteristic changes in factory operation data, violation behavior images, and equipment degradation health indicators, and generating a factory text report based on the textual descriptions; Collect text data on factory equipment operation, maintenance, and production management to build an industry domain corpus; Use the pre-trained language model BERT and train it on the domain corpus to learn industry-specific knowledge and language habits; The output text of the feature change detection, violation detection and equipment health indicator prediction models is used as the input of the report generation model to design a framework for multi-document summary generation. A hierarchical attention mechanism is introduced into the framework to automatically learn how to aggregate information at different text granularities, including words, sentences, and paragraphs. The Transformer self-attention mechanism model is used as the decoder to generate the factory operation report.

4. The industrial equipment maintenance method based on multi-source heterogeneous data according to claim 3 is characterized in that: The industrial equipment maintenance method further includes: defining entity types in the factory field, using a conditional random field (CRF) sequence labeling model to identify entities in the text; Define the relationship types in the factory field, use the PCNN relationship extraction model based on neural network to extract the relationship between entities from the text; deduplicate and merge the identified entities and relationships to eliminate redundant information; use external knowledge base to disambiguate and complete entities and relationships; embed entities and relationships into a low-dimensional semantic space through the knowledge representation learning method TransE; convert the extracted entities and relationships into nodes and edges of the knowledge graph; use a graph database to store and manage the knowledge graph.

5. An industrial equipment maintenance platform based on multi-source heterogeneous data, which is used to implement the industrial equipment maintenance method based on multi-source heterogeneous data according to any one of claims 1 to 4, characterized in that: include: Data acquisition module, equipment health quantification module, data feature extraction module and violation detection module; The data acquisition module is used to acquire multi-source heterogeneous data, wherein the multi-source heterogeneous data includes industrial equipment operation and maintenance data, industrial equipment monitoring data, factory operation data and monitoring image data; The equipment health quantification module generates an industrial equipment degradation health index based on the industrial equipment operation and maintenance data, wherein the equipment degradation health index includes an equipment efficiency decline rate and a failure rate increase rate, and the equipment degradation health index is used to quantitatively indicate the equipment health status; The data feature extraction module is used to extract features from the acquired plant operation data and generate feature changes of the plant operation data; The violation detection module is used to construct a convolutional neural network model, input the monitoring image data into the convolutional neural network model, and output a violation image.

6. An electronic device, characterized in that: include: A processor and a memory, wherein the memory stores a computer program that can be called by the processor; the processor executes the industrial equipment maintenance method based on multi-source heterogeneous data as described in any one of claims 1 to 4 by calling the computer program stored in the memory.

7. A computer-readable storage medium, characterized in that: Instructions are stored, and when the instructions are executed on a computer, the computer is caused to execute the industrial equipment maintenance method based on multi-source heterogeneous data as described in any one of claims 1 to 4.

Images