Automatic production equipment remote monitoring and fault diagnosis method and system based on industrial internet

Abstract

The invention discloses an industrial internet-based automatic production equipment remote monitoring and fault diagnosis method and system, and relates to the technical field of intelligent monitoring and fault diagnosis. By integrating multiple types of high-precision sensors and a data fusion technology, the comprehensiveness and the high efficiency of data acquisition and processing are remarkably improved, the problem that the data dimension of a traditional method is limited is solved, meanwhile, data transmission is optimized through edge computing nodes and a low-delay communication protocol, the reliability and the real-time performance of data transmission are enhanced, and the data transmission efficiency is improved. Besides, the data set is dynamically adjusted and optimized through deep learning and multiple intelligent algorithms, the intelligent level and adaptability of fault diagnosis are greatly improved, and the system can efficiently and accurately operate in a complex industrial scene.

Description

Technical Field

[0001] The present invention relates to the field of intelligent monitoring and fault diagnosis technology, and specifically to a method and system for remote monitoring and fault diagnosis of automated production equipment based on the Industrial Internet. Background Art

[0002] As a key technology field in promoting the digital transformation of the manufacturing industry, the Industrial Internet plays an irreplaceable role in realizing remote monitoring and fault diagnosis of automated production equipment. With the in-depth development of intelligent manufacturing, real-time understanding of equipment operating status and accurate prediction of faults are directly related to production efficiency and economic benefits. However, current solutions still have significant limitations in practical applications.

[0003] Traditional methods mostly rely on a single type of sensor to collect data, which has limited data dimensions and cannot fully reflect the status of the equipment. At the same time, data transmission is limited by bandwidth and latency, and the processing process lacks efficient algorithm support, resulting in insufficient accuracy and real-time performance of fault diagnosis. These defects are particularly evident in complex industrial scenarios, and breakthrough technologies are urgently needed to fill the gap. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for remote monitoring and fault diagnosis of automated production equipment based on the Industrial Internet.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] This application provides a remote monitoring and fault diagnosis method for automated production equipment based on the Industrial Internet, comprising the following steps:

[0007] By integrating multiple types of high-precision sensors to obtain multi-source heterogeneous data on equipment operation, and using data fusion technology to pre-process the collected multi-dimensional data, we can obtain the first data set that fully reflects the equipment status;

[0008] Obtaining device status reflection information from the first data set, locally compressing the first data set using an edge computing node, and transmitting the compressed first data set to the cloud via a low-latency communication protocol to obtain an optimized second data set;

[0009] After receiving the optimized second data set in the cloud, a distributed computing framework is used to perform parallel decompression and feature extraction on the second data set to obtain a third data set containing key features of device operation;

[0010] Extracting fault-related features from the third data set, training and optimizing the third data set through a pre-established deep learning model to generate a fourth data set with preliminary intelligent capabilities;

[0011] When the fault prediction accuracy of the fourth dataset is determined to be lower than a preset threshold in the simulation environment, the fourth dataset is iteratively updated using an incremental learning algorithm and combined with the real-time collected environmental change data to obtain a fifth dataset that can be dynamically adjusted;

[0012] The improvement direction of the weak adaptive ability is obtained from the fifth data set, and the fifth data set is optimized online using the reinforcement learning algorithm to generate the sixth data set that can adapt to changes in the equipment operating environment.

[0013] Furthermore, a first data set that fully reflects the device status is obtained, specifically including:

[0014] The multi-source heterogeneous data collected by high-precision sensors during device operation is obtained by adopting multi-channel parallel processing technology to obtain an initial collection data set. The initial collection data set is then pre-processed using data fusion technology, and the multi-dimensional data is integrated using a weighted average algorithm to obtain a first data set.

[0015] According to the heterogeneous data processing requirements in the first data set, determine whether the data dimension exceeds a preset threshold. If so, reduce the dimension using a principal component analysis algorithm to obtain a reduced-dimensional data set.

[0016] Based on the reduced-dimensional dataset, the feature vector of the equipment's operating status is obtained. Key indicators of the status are determined through feature extraction technology. When the key indicators deviate from the normal range, the feature vector is classified through a clustering algorithm to determine the abnormal points of equipment operation.

[0017] Through the mapping relationship between abnormal points and comprehensive reflection of the status, the equipment status analysis results are obtained, the status distribution characteristics are obtained, and then the time series analysis tools are used to determine the dynamic change trend of equipment operation.

[0018] Further, obtaining an optimized second data set specifically includes: acquiring device status information from the first data set through a sensor, extracting data using a preset sampling frequency to obtain initial state data, performing local compression on the initial state data at an edge computing node, reducing data redundancy using an LZ77 algorithm, and generating a second data set;

[0019] The second data set is transmitted to the cloud via a low-latency communication protocol, and the MQTT protocol is used for data distribution. After confirming that the transmission is complete, the transmission timestamp is extracted from the second data set. When the transmission time is lower than a preset threshold, the efficiency is determined to be met, and the transmission performance index is obtained.

[0020] The integrity of the data received by the cloud is analyzed based on the transmission performance indicators, and the data consistency is verified using the verification method to determine the data quality. The compression parameters of the edge computing node are adjusted according to the data quality results to generate an optimized second data set.

[0021] Furthermore, a third data set containing key characteristics of equipment operation is obtained, specifically including:

[0022] receiving an optimized second data set through the cloud, obtaining a complete data stream from a storage system, and performing sharding processing on the second data set using a distributed computing framework to obtain multiple parallel task units;

[0023] Then, the parallel task units are decompressed to obtain the decompressed raw data content. The raw data content is processed using a feature extraction algorithm to obtain a feature set containing the running status. When the running status in the feature set exceeds a preset threshold, the feature set is grouped using a clustering algorithm to determine the abnormal feature subset.

[0024] According to the mapping relationship between the abnormal feature subset and the equipment operation, the distribution law of the key features is determined, and then the regression algorithm is used to fit the distribution law of the key features to obtain a prediction model for the equipment operation trend.

[0025] Furthermore, a fourth data set with preliminary intelligent capabilities is generated, specifically including:

[0026] Obtaining initial fault-related features from the third data set, processing them using feature extraction technology to obtain a first feature set, and then training them using a pre-established prediction model for equipment operation trends to obtain a preliminary training model;

[0027] The first feature set is optimized through the preliminary training model to generate a second feature set. When the intelligence capability of the second feature set is lower than a preset threshold, the training process is adjusted through the optimization method to obtain a third feature set.

[0028] Based on the third feature set, a fourth dataset with preliminary intelligent capabilities is constructed using data generation technology. The fourth dataset is used to verify the training effect of the deep learning model and determine whether the intelligent capabilities meet the requirements.

[0029] After obtaining the verification results, the fourth data set is iteratively processed using the adjusted optimization method to obtain the final data set.

[0030] Furthermore, a fifth data set capable of dynamic adjustment is obtained, specifically including:

[0031] Fault prediction is performed on the fourth data set in a simulation environment to obtain accuracy data. When the accuracy is lower than a preset threshold, the fourth data set is iteratively updated using an incremental learning algorithm to obtain an updated data set.

[0032] Collect environmental change data in real time to obtain external input data. Then, through data combination methods, fuse the updated data set with the external input data to obtain a preliminary adjusted data set.

[0033] Dynamic adjustment technology is used to optimize the preliminary adjusted data set to obtain the fifth data set. Fault prediction features are extracted from the fifth data set to determine whether the preset threshold requirements are met. When the fault prediction accuracy of the fifth data set is still lower than the preset threshold, the iterative update of the incremental learning algorithm is repeated to obtain the final adjusted data set.

[0034] Furthermore, a sixth data set is generated that can adapt to changes in the device operating environment, specifically including:

[0035] Obtaining data extraction results from the fifth data set, using preset rules to determine features with weak adaptability, obtaining a set of improvement directions, and extracting key features from the set of improvement directions;

[0036] The fifth dataset is optimized online using a reinforcement learning algorithm to generate preliminary optimized data. It then determines whether the environmental changes exceed a preset threshold. If so, the optimization parameters are adjusted to obtain data with enhanced adaptability.

[0037] Based on the data with enhanced adaptability, the environmental change trend is judged and the initial version of the sixth data set is generated. Then, data extraction technology is used to separate the device operation-related features to obtain the environmental adaptation feature set. At the same time, the optimization process is applied to adjust the data structure to generate the final sixth data set.

[0038] The method further includes: determining a prediction model adjustment plan based on the sixth data set, and performing adaptive training on the sixth data set across device scenarios using a transfer learning algorithm to obtain a final seventh data set;

[0039] It also includes: Based on the seventh data set, the prediction model is deployed to the edge device through low-latency communication technology, and the device status feedback data is obtained from the edge device in real time to determine the operating stability of the remote monitoring system.

[0040] This application provides an industrial internet-based automated production equipment remote monitoring and fault diagnosis system, which is used to implement an industrial internet-based automated production equipment remote monitoring and fault diagnosis method, including:

[0041] The data acquisition and preprocessing module integrates multiple types of high-precision sensors to acquire multi-source heterogeneous data on equipment operation, and uses data fusion technology to preprocess the collected multi-dimensional data to obtain the first data set that fully reflects the equipment status;

[0042] The data transmission and optimization module obtains device status information from the first data set, locally compresses the first data set using edge computing nodes, transmits the compressed first data set to the cloud via a low-latency communication protocol, obtains an optimized second data set, and determines the efficiency of the transmission process;

[0043] A data processing and feature extraction module receives the optimized second data set from the cloud and uses a distributed computing framework to perform parallel decompression and feature extraction on the second data set to obtain a third data set containing key features of device operation;

[0044] An intelligent fault diagnosis module extracts fault-related features from the third data set, trains and optimizes the third data set using a pre-established deep learning model, and generates a fourth data set with preliminary intelligent capabilities;

[0045] A dynamic adjustment and optimization module, which iteratively updates the fourth dataset using an incremental learning algorithm when the fault prediction accuracy of the fourth dataset in a simulated environment is determined to be lower than a preset threshold, and combines it with the real-time collected environmental change data to obtain a fifth dataset that can be dynamically adjusted;

[0046] An adaptive optimization module obtains improvement directions for weak adaptability from the fifth dataset, performs online optimization on the fifth dataset using a reinforcement learning algorithm, and generates a sixth dataset that can adapt to changes in the device operating environment;

[0047] A cross-device scenario adaptation module determines a prediction model adjustment plan based on the sixth data set, and uses a transfer learning algorithm to perform adaptive training on the sixth data set for cross-device scenarios to obtain a final seventh data set;

[0048] The remote monitoring and evaluation module, based on the seventh data set, deploys the prediction model to the edge device through low-latency communication technology, obtains device status feedback data from the edge device in real time, and judges the operational stability of the remote monitoring system.

[0049] The beneficial effects of the present invention are:

[0050] This invention integrates multiple types of high-precision sensors to collect multi-source heterogeneous data on equipment operation, and uses data fusion technology for preprocessing to generate a first data set that comprehensively reflects the equipment status. This process not only solves the problem of traditional methods relying on a single sensor and limited data dimensions, but also integrates multidimensional data through multi-channel parallel processing technology and weighted averaging algorithms, improving the comprehensiveness of data collection. At the same time, the principal component analysis algorithm is used to reduce the dimensionality of high-dimensional data, reducing computational complexity and retaining key features. This provides a richer and more accurate data foundation for subsequent fault diagnosis and significantly improves the overall performance of the system.

[0051] The present invention utilizes edge computing nodes to perform local compression processing on initial state data, adopts the LZ77 algorithm to reduce data redundancy, and efficiently transmits the compressed data to the cloud through a low-latency communication protocol (such as MQTT). This process effectively solves the problem of data transmission being limited by bandwidth and latency in traditional methods. By dynamically adjusting the compression parameters of edge computing nodes, the integrity and consistency of data during transmission are ensured, further improving the reliability of data transmission. At the same time, the application of low-latency communication protocols ensures the real-time performance of data transmission, enabling the system to quickly respond to changes in device status, providing strong support for real-time monitoring and fault diagnosis.

[0052] The present invention trains and optimizes the third data set through a deep learning model to generate a fourth data set with preliminary intelligent capabilities, and uses incremental learning algorithms, reinforcement learning algorithms, and transfer learning algorithms to dynamically adjust and optimize the data set. This process significantly improves the intelligence level of fault diagnosis and its adaptability to environmental changes. The incremental learning algorithm can dynamically adjust the model based on real-time collected environmental change data, the reinforcement learning algorithm further enhances the model's adaptive capabilities, enabling it to better respond to changes in the equipment operating environment, and the application of the transfer learning algorithm enables the model to quickly adapt to different equipment scenarios, improving the versatility and reliability of the system. The present invention not only improves the accuracy of fault diagnosis, but also enhances the system's intelligence level, enabling it to operate efficiently in complex industrial scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] For better understanding and implementation, the technical solution of the present application is described in detail below with reference to the accompanying drawings.

[0054] Figure 1 A flow chart of a method for remote monitoring and fault diagnosis of automated production equipment based on the Industrial Internet provided in Example 1 of the present application;

[0055] Figure 2 A schematic diagram of a process for generating a sixth data set for the method for remote monitoring and fault diagnosis of automated production equipment based on the Industrial Internet provided in Example 1 of the present application;

[0056] Figure 3 A structural diagram of the remote monitoring and fault diagnosis system for automated production equipment based on the Industrial Internet provided in Example 2 of the present application. DETAILED DESCRIPTION

[0057] To further illustrate the technical means and effectiveness of the present invention in achieving its intended purpose, exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present application. Rather, they are merely examples of methods and systems consistent with certain aspects of the present application, as detailed in the appended claims.

[0058] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. As used in this application and the appended claims, the singular forms "a," "an," "the," and "the" are intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0059] The following describes in detail the specific implementation methods, features and effects of the present invention in conjunction with the accompanying drawings and preferred embodiments.

[0060] Example 1

[0061] See also Figure 1-Figure 2 This embodiment provides a remote monitoring and fault diagnosis method for automated production equipment based on the Industrial Internet, comprising the following steps:

[0062] S1. To address the issues of limited data dimensions and comprehensive data collection, we integrate multiple types of high-precision sensors to obtain multi-source heterogeneous data on equipment operation. We use data fusion technology to pre-process the collected multi-dimensional data to obtain the first data set that fully reflects the equipment status.

[0063] Furthermore, in step S1, a first data set that fully reflects the device status is obtained, specifically including:

[0064] The multi-source heterogeneous data collected by high-precision sensors during device operation is obtained by adopting multi-channel parallel processing technology to obtain an initial collection data set. The initial collection data set is then pre-processed using data fusion technology, and the multi-dimensional data is integrated using a weighted average algorithm to obtain a first data set.

[0065] According to the heterogeneous data processing requirements in the first data set, determine whether the data dimension exceeds a preset threshold. If so, reduce the dimension using a principal component analysis algorithm to obtain a reduced-dimensional data set.

[0066] Based on the reduced-dimensional dataset, the feature vector of the equipment's operating status is obtained. Key indicators of the status are determined through feature extraction technology. When the key indicators deviate from the normal range, the feature vector is classified through a clustering algorithm to determine the abnormal points of equipment operation.

[0067] Through the mapping relationship between abnormal points and comprehensive reflection of the status, the equipment status analysis results are obtained, the status distribution characteristics are obtained, and then the time series analysis tools are used to determine the dynamic change trend of equipment operation.

[0068] Specifically, high-precision sensors are used to collect multi-source, heterogeneous data during device operation. The key is to obtain comprehensive information about the device's operation. For example, when an industrial motor is running, vibration sensors, temperature sensors, and current sensors can be deployed to collect amplitude, temperature, and current values, respectively, forming a multi-source data stream. For example, vibration data may be recorded at a frequency of 1,000 times per second, temperature data may be updated once a minute, and current data may change in real time. This heterogeneity requires parallel processing.

[0069] Multi-channel parallel processing technology can synchronously input these data streams into the processing unit. For example, a multi-threaded processor can parse each type of data separately, ultimately integrating it into an initial collected data set containing timestamps and corresponding multidimensional values. When using data fusion technology to preprocess the initial collected data set, the weighted average algorithm is key. For example, assuming the vibration data weight is 0.5, the temperature is 0.3, and the current is 0.2, after assigning weights based on the equipment's operating characteristics, the weighted average is calculated to obtain the first data set. In one possible implementation, if the vibration value at a certain moment is 10mm / s, the temperature is 50°C, and the current is 5A, the fusion value is 10×0.5+50×0.3+5×0.2=21. This method can smooth noise, highlight key trends, and improve data consistency. For the first data set, if the data dimension exceeds a preset threshold, such as a dimension greater than 10, it is necessary to reduce the dimension using the principal component analysis algorithm.

[0070] Specifically, assuming the first dataset contains 12-dimensional data such as vibration, temperature, current, and noise, principal component analysis can extract the first three principal components, retaining 90% of the information, resulting in a reduced-dimensional dataset. This dimensionality reduction reduces computational complexity while preserving key features. For example, the reduced-dimensional data may focus on the combined effects of vibration and temperature, reflecting the core state of the equipment. When extracting feature vectors from the reduced-dimensional dataset, it is understandable that feature extraction techniques focus on key state indicators. For example, statistical analysis can be used to calculate the mean and variance of vibration and the peak value of temperature, generating feature vectors [5, 2, 60], representing the vibration mean, variance, and peak value of temperature, respectively. If the normal range is [4-6, 1-3, 50-70], but a particular detection value is [8, 4, 75], it deviates from the normal range and requires further analysis. This method intuitively reflects the equipment state and facilitates anomaly detection. If the key state indicators are abnormal, the feature vectors are classified using a clustering algorithm.

[0071] Preferably, K-means clustering is used to classify feature vectors into "normal" and "abnormal" categories. For example, after clustering feature vectors from multiple tests, outliers are concentrated near [8, 4, 75], indicating a possible equipment overload. It should be noted that clustering results can also be combined with historical data to enhance classification accuracy. When analyzing results by mapping outliers to states, in one embodiment, a state distribution feature can be constructed. For example, if the outlier mapping shows a correlation between increased temperature and increased vibration, the state distribution feature might be "high temperature and high vibration," which helps locate the cause of the fault. Based on the state distribution feature, time series analysis tools can predict dynamic trends. For example, based on the temperature and vibration data from the past 24 hours, the moving average method can be used to detect a continuous temperature increase and predict that the temperature may exceed 80°C in 12 hours. This prediction can provide early warning, optimize maintenance plans, and improve equipment reliability.

[0072] S2. Obtain device status information from the first dataset. Using edge computing nodes to locally compress the first dataset to account for transmission delay and bandwidth limitations, transmit the compressed first dataset to the cloud via a low-latency communication protocol to obtain an optimized second dataset, and determine the efficiency of the transmission process.

[0073] Furthermore, in step S2, obtaining an optimized second data set specifically includes: acquiring device status information from the first data set through a sensor, extracting data using a preset sampling frequency to obtain initial state data, performing local compression on the initial state data at an edge computing node, and using an LZ77 algorithm to reduce data redundancy to generate a second data set;

[0074] The second data set is transmitted to the cloud via a low-latency communication protocol, and the MQTT protocol is used for data distribution. After confirming that the transmission is complete, the transmission timestamp is extracted from the second data set. When the transmission time is lower than a preset threshold, the efficiency is determined to be met, and the transmission performance index is obtained.

[0075] The integrity of the data received by the cloud is analyzed based on the transmission performance indicators, and the data consistency is verified using the verification method to determine the data quality. The compression parameters of the edge computing node are adjusted according to the data quality results to generate an optimized second data set.

[0076] Specifically, the entire process of data collection, compression, transmission, and quality control has been optimized. Specific results include: significantly reducing data redundancy, lowering transmission latency, and improving the efficiency and reliability of data transmission; further improving data quality by dynamically adjusting compression parameters, ensuring the integrity and consistency of data received in the cloud; and ultimately, the generated optimized second data set provides a more accurate and reliable basis for equipment status monitoring and fault diagnosis, enhancing the performance and stability of the entire system.

[0077] S3. After receiving the optimized second dataset in the cloud, to address the problem of low processing efficiency, a distributed computing framework is used to perform parallel decompression and feature extraction on the second dataset to obtain a third dataset containing key features of device operation;

[0078] Furthermore, in step S3, a third data set containing key characteristics of device operation is obtained, specifically including:

[0079] receiving an optimized second data set through the cloud, obtaining a complete data stream from a storage system, and performing sharding processing on the second data set using a distributed computing framework to obtain multiple parallel task units;

[0080] Then, the parallel task units are decompressed to obtain the decompressed raw data content. The raw data content is processed using a feature extraction algorithm to obtain a feature set containing the running status. When the running status in the feature set exceeds a preset threshold, the feature set is grouped using a clustering algorithm to determine the abnormal feature subset.

[0081] According to the mapping relationship between the abnormal feature subset and the equipment operation, the distribution law of the key features is determined, and then the regression algorithm is used to fit the distribution law of the key features to obtain a prediction model for the equipment operation trend.

[0082] Specifically, by adopting a distributed computing framework to perform parallel decompression and feature extraction on the optimized second data set, the data processing efficiency was significantly improved, and the third data set containing the key characteristics of equipment operation was quickly obtained. Further, through anomaly detection and cluster analysis of the feature set, the abnormal feature subset was accurately identified, and the regression algorithm was used to fit the distribution law of key features, and a prediction model for equipment operation trends was established, which realized real-time monitoring and trend prediction of equipment operation status, provided strong support for fault warning and maintenance decision-making, and improved the level of intelligent equipment management.

[0083] S4. To address the problem of low intelligence level of the fault diagnosis model, extract fault-related features from the third data set, train and optimize the third data set through a pre-established deep learning model, and generate a fourth data set with preliminary intelligence capabilities;

[0084] Furthermore, in step S4, a fourth data set with preliminary intelligent capabilities is generated, specifically including:

[0085] Obtaining initial fault-related features from the third data set, processing them using feature extraction technology to obtain a first feature set, and then training them using a pre-established prediction model for equipment operation trends to obtain a preliminary training model;

[0086] The first feature set is optimized through the preliminary training model to generate a second feature set. When the intelligence capability of the second feature set is lower than a preset threshold, the training process is adjusted through the optimization method to obtain a third feature set.

[0087] Based on the third feature set, a fourth dataset with preliminary intelligent capabilities is constructed using data generation technology. The fourth dataset is used to verify the training effect of the deep learning model and determine whether the intelligent capabilities meet the requirements.

[0088] After obtaining the verification results, the fourth data set is iteratively processed using the adjusted optimization method to obtain the final data set.

[0089] Specifically, by extracting fault-related features from the third data set and using a deep learning model for training and optimization, a fourth data set with preliminary intelligent capabilities was generated. This process not only improved the intelligence level of the fault diagnosis model, but also further enhanced the model's ability to identify and predict fault characteristics through feature optimization, model adjustment, and data generation technology. The final data set obtained through iterative processing can more accurately reflect the equipment failure situation, provide strong support for intelligent fault diagnosis, and significantly improve the efficiency and accuracy of fault diagnosis.

[0090] S5. When the fault prediction accuracy of the fourth dataset is determined to be lower than a preset threshold in the simulation environment, the fourth dataset is iteratively updated using an incremental learning algorithm and combined with the real-time collected environmental change data to obtain a fifth dataset that can be dynamically adjusted.

[0091] Furthermore, in step S5, a fifth data set capable of dynamic adjustment is obtained, which specifically includes:

[0092] Fault prediction is performed on the fourth data set in a simulation environment to obtain accuracy data. When the accuracy is lower than a preset threshold, the fourth data set is iteratively updated using an incremental learning algorithm to obtain an updated data set.

[0093] Collect environmental change data in real time to obtain external input data. Then, through data combination methods, fuse the updated data set with the external input data to obtain a preliminary adjusted data set.

[0094] Dynamic adjustment technology is used to optimize the preliminary adjusted data set to obtain the fifth data set. Fault prediction features are extracted from the fifth data set to determine whether the preset threshold requirements are met. When the fault prediction accuracy of the fifth data set is still lower than the preset threshold, the iterative update of the incremental learning algorithm is repeated to obtain the final adjusted data set.

[0095] Specifically, fault prediction is performed on the fourth data set through a simulated environment. When the accuracy is lower than the preset threshold, an incremental learning algorithm is used for iterative updates. Combined with the real-time collected environmental change data, dynamic adjustment and optimization of the data set are achieved. This process not only improves the accuracy of fault prediction, but also enhances the model's adaptability to environmental changes. The final fifth data set can more accurately reflect equipment failure conditions, provide more reliable data support for fault diagnosis, and significantly improve the system's intelligence level and dynamic adaptability.

[0096] S6. Determine improvement directions for areas with weak adaptability from the fifth dataset, and use a reinforcement learning algorithm to perform online optimization on the fifth dataset based on the impact of environmental changes, thereby generating a sixth dataset that can adapt to changes in the device operating environment.

[0097] Furthermore, in step S6, a sixth data set capable of adapting to changes in the operating environment of the self-adapting device is generated, specifically including:

[0098] S61. Obtain data extraction results from the fifth data set, use preset rules to determine features with weak adaptability, obtain a set of improvement directions, and extract key features from the set of improvement directions;

[0099] S62. Perform online optimization on the fifth data set using a reinforcement learning algorithm to generate preliminary optimized data. Then, determine whether the environmental change exceeds a preset threshold. If so, adjust the optimization parameters to obtain data with enhanced adaptability.

[0100] S63. Based on the data with enhanced adaptability, determine the trend of environmental changes and generate an initial version of the sixth data set. Then, use data extraction technology to separate device operation-related features to obtain an environmental adaptation feature set. At the same time, apply an optimization process to adjust the data structure and generate the final sixth data set.

[0101] Specifically, when obtaining data extraction results through the fifth data set, it can be understood that the core of data extraction is to screen out meaningful features from complex data sets. For example, in the field of equipment failure prediction, assuming that the fifth data set contains information such as temperature, vibration frequency, and current value during equipment operation, through data extraction technology, specific records of abnormal temperature fluctuations and excessive vibration frequencies can be obtained. For these extraction results, preset rules are used to judge features with weak adaptive capabilities.

[0102] For example, the preset rule can be set as "a feature change exceeding 20% ​​is considered to have insufficient adaptability." Then, it is found that the temperature feature frequently exceeds the threshold when the environment changes suddenly, indicating that its adaptability is weak, and it is included in the improvement direction set. When extracting key features from the improvement direction set, specifically, the focus can be on the temperature feature because it is directly related to the stability of equipment operation.

[0103] In one possible implementation, a reinforcement learning algorithm is used to perform online optimization on the fifth data set. For example, reinforcement learning can simulate the operating state of the equipment in a high-temperature environment, gradually adjust the data weights, and generate preliminary optimized data. Assume that the temperature feature accounts for 30% in the initial data, and after reinforcement learning, it is increased to 50% to enhance its contribution to fault prediction. The advantage of this online optimization is that it can respond to changes in equipment status in real time.

[0104] When analyzing the equipment operating status based on preliminary optimization data, it should be noted that if the environmental changes exceed the preset threshold, for example, the external temperature suddenly rises from 25 degrees to 40 degrees, the optimization parameters are adjusted. Preferably, the weight of the environmental variables can be increased, such as adjusting the humidity impact from 10% to 20%, to obtain data with enhanced adaptability. This adjustment can enable the data to better reflect the actual operating conditions. For example, data with enhanced adaptability can more accurately predict the risk of equipment failure under high temperatures.

[0105] When judging the trend of environmental changes based on data with enhanced adaptability, in one embodiment, the temperature and vibration change curves can be analyzed to predict abnormal peaks that may occur in the next 24 hours, thereby generating an initial version of the sixth data set. The advantage of this prediction is that it provides a direction for subsequent optimization. When using data extraction technology to separate equipment operation-related features, for example, the periodic change characteristics of the vibration frequency can be extracted from the initial version to obtain an environmental adaptation feature set. Assuming that the vibration frequency shows abnormal fluctuations every 2 hours under high temperature, this feature set can clearly reflect the environmental impact.

[0106] When applying the optimization process to the environmental adaptation feature set to adjust the data structure, specifically, by reorganizing the data priority, the vibration frequency and temperature features can be placed in the forefront to generate the final sixth data set. For example, in the adjusted data structure, the sampling interval of the vibration frequency is shortened from 10 minutes to 5 minutes to improve real-time performance. The beneficial effect of this optimization is that the data can more accurately match the equipment operation requirements and provide a more reliable basis for fault prediction.

[0107] S7. Determine a prediction model adjustment plan based on the sixth data set. To improve the reliability of the remote monitoring system, use a transfer learning algorithm to perform adaptive training on the sixth data set across device scenarios to obtain a final seventh data set.

[0108] Furthermore, in step S7, a final seventh data set is obtained, which specifically includes:

[0109] Extracting feature data from the sixth data set, determining an adjustment plan for the prediction model using statistical analysis to obtain an optimization direction, obtaining parameter configuration based on the optimization direction, and preprocessing the sixth data set using a transfer learning algorithm to obtain a first intermediate data set;

[0110] Performing feature mapping across device scenarios on the first intermediate dataset, and adjusting the model structure using a transfer learning algorithm to obtain a second intermediate dataset;

[0111] Adaptive training is performed on the second intermediate data set, and a convergence state during the training process is determined to obtain a third intermediate data set. When the third intermediate data set meets a preset threshold, model parameters are optimized using information fusion technology to obtain a fourth intermediate data set.

[0112] The scenario adaptation is verified based on the fourth intermediate data set, and the cross-device performance is evaluated using a test tool to obtain the fifth intermediate data set, and then the seventh data set is output using data integration technology.

[0113] Specifically, when extracting feature data from the sixth dataset, it can be understood that feature extraction is a fundamental step in the entire process. For example, assuming that the equipment operating data includes variables such as temperature, pressure, and operating time, statistical methods such as mean analysis or variance calculation can be used to extract features that are sensitive to environmental changes. For example, the temperature fluctuation range may be a key feature. Statistical analysis shows that its mean is 25 degrees, but the variance reaches 10, indicating high environmental instability. This feature can serve as a basis for adjusting the prediction model, and the resulting optimization direction may be to enhance the model's robustness to temperature fluctuations.

[0114] In one possible implementation, when using statistical analysis to determine adjustments to the prediction model, the correlation between features can be analyzed based on historical data. For example, if temperature and pressure are positively correlated, and the statistical results show a correlation coefficient of 0.8, the adjustment plan may tend to jointly optimize the weights of these two variables. When obtaining parameter configuration from the optimization direction, specifically, the temperature weight can be set to 0.6 and the pressure weight can be set to 0.4 to form an initial parameter set. When the transfer learning algorithm preprocesses the sixth data set, it is preferably possible to leverage knowledge from existing device models to migrate the parameter configuration to the new scenario. For example, migrating the temperature prediction model of the old device to the new device. After preprocessing, a first intermediate data set is obtained, whose data structure has been initially adapted to the new environment.

[0115] When performing feature mapping across device scenarios for the first intermediate dataset, it should be noted that feature mapping needs to take into account the differences between devices. For example, the temperature sensor of the new device has higher accuracy. During mapping, the temperature range of the old data can be adjusted from 20-30 degrees to 22-28 degrees to match the actual operating range of the new device. After adjusting the model structure using the transfer learning algorithm, the model of the second intermediate dataset may pay more attention to the fitting ability of high-precision data. When performing adaptive training using the second intermediate dataset, for example, the number of training iterations can be set to 1000 times to observe whether the loss value tends to stabilize. If the convergence state is good, the feature distribution of the third intermediate dataset will be closer to the actual operating environment. If the third intermediate dataset meets the preset threshold, for example, the loss value is lower than 0.05, the model parameters are optimized through information fusion technology.

[0116] In one embodiment, the characteristic vectors of temperature and pressure can be weightedly fused to generate a more stable parameter set to obtain a fourth intermediate data set. When performing scene adaptation verification based on the fourth intermediate data set, specifically, a test tool can be used to simulate the operation of the device at a high temperature of 35 degrees and a low temperature of 15 degrees to evaluate cross-device performance. If performance indicators such as prediction accuracy reach 90%, the fifth intermediate data set can be regarded as a reliable intermediate result. When the final result is generated by the fifth intermediate data set, it can be understood that the data integration technology needs to ensure the integrity of the output data. For example, the predicted values ​​of temperature, pressure and operating time are integrated into time series data to form a seventh data set. This method can effectively support real-time monitoring of the operating status of the equipment.

[0117] In one embodiment, the seventh data set can be directly used for equipment maintenance scheduling and early warning of potential failures. This process progresses step by step from feature extraction to final output, ensuring the adaptability of the data to the equipment environment and helping to improve the stability and predictive capabilities of the system.

[0118] S8. Based on the seventh data set, the prediction model is deployed to the edge device through low-latency communication technology, and the device status feedback data is obtained from the edge device in real time to determine the operating stability of the remote monitoring system.

[0119] Furthermore, in step S8, a first data set that fully reflects the device status is obtained, specifically including: optimizing the prediction model using a random forest algorithm, deploying the optimized prediction model to the edge device, and obtaining a deployment completion signal through low-latency communication technology;

[0120] Based on the state feedback data generated by the edge device during operation, a real-time acquisition technology is used to obtain a feedback data set, which is then processed using a mean filtering method to obtain a smoothed data set.

[0121] Based on the smoothed data set, the support vector machine algorithm is used to determine the operation stability trend. The stability deviation value is obtained by comparing the operation stability trend with the preset standard of the remote monitoring system.

[0122] When the stability deviation value exceeds the preset range, an adjustment instruction is sent to the edge device through communication technology to obtain updated status feedback data.

[0123] Specifically, by optimizing the prediction model and deploying it to edge devices, combined with low-latency communication technology to obtain real-time device status feedback data, the system achieves efficient monitoring and dynamic adjustment of the remote monitoring system's operational stability. This process utilizes the random forest algorithm to optimize the prediction model, improving its accuracy and robustness. Feedback data is processed using a mean filter to eliminate noise, making the data smoother and more reliable. A support vector machine algorithm is then used to determine operational stability trends and accurately assess system stability deviations. When the deviation exceeds a preset range, adjustment instructions are promptly sent to the edge device to ensure stable system operation, significantly improving the reliability and adaptability of the remote monitoring system.

[0124] Example 2

[0125] See also Figure 3 This embodiment provides an industrial Internet-based automated production equipment remote monitoring and fault diagnosis system, which is used to implement an industrial Internet-based automated production equipment remote monitoring and fault diagnosis method, including:

[0126] The data acquisition and preprocessing module integrates multiple types of high-precision sensors to acquire multi-source heterogeneous data on equipment operation, and uses data fusion technology to preprocess the collected multi-dimensional data to obtain the first data set that fully reflects the equipment status;

[0127] The data transmission and optimization module obtains device status information from the first data set, uses edge computing nodes to locally compress the first data set to address transmission delay and bandwidth limitations, and transmits the compressed first data set to the cloud via a low-latency communication protocol to obtain an optimized second data set, thereby determining the efficiency of the transmission process.

[0128] The data processing and feature extraction module receives the optimized second dataset from the cloud and, to address the problem of low processing efficiency, uses a distributed computing framework to perform parallel decompression and feature extraction on the second dataset to obtain a third dataset containing key features of device operation;

[0129] An intelligent fault diagnosis module extracts fault-related features from the third data set, trains and optimizes the third data set using a pre-established deep learning model, and generates a fourth data set with preliminary intelligent capabilities;

[0130] A dynamic adjustment and optimization module, which iteratively updates the fourth dataset using an incremental learning algorithm when the fault prediction accuracy of the fourth dataset in a simulated environment is determined to be lower than a preset threshold, and combines it with the real-time collected environmental change data to obtain a fifth dataset that can be dynamically adjusted;

[0131] The adaptive optimization module obtains improvement directions for weak adaptability from the fifth dataset, and uses a reinforcement learning algorithm to perform online optimization on the fifth dataset based on the impact of environmental changes, thereby generating a sixth dataset that can adapt to changes in the device operating environment.

[0132] The cross-device scenario adaptation module determines the prediction model adjustment plan based on the sixth data set. To improve the reliability of the remote monitoring system, the sixth data set is adaptively trained across device scenarios using a transfer learning algorithm to obtain the final seventh data set.

[0133] The remote monitoring and evaluation module, based on the seventh data set, deploys the prediction model to the edge device through low-latency communication technology, obtains device status feedback data from the edge device in real time, and judges the operational stability of the remote monitoring system.

[0134] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as above in terms of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can, without departing from the scope of the technical solution of the present invention, make some changes or modifications to equivalent embodiments using the technical contents disclosed above. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A remote monitoring and fault diagnosis method for automated production equipment based on the Industrial Internet, characterized by: The steps include: By integrating multiple types of high-precision sensors to obtain multi-source heterogeneous data on equipment operation, and using data fusion technology to pre-process the collected multi-dimensional data, we can obtain the first data set that fully reflects the equipment status; Obtaining device status reflection information from the first data set, locally compressing the first data set using an edge computing node, and transmitting the compressed first data set to the cloud via a low-latency communication protocol to obtain an optimized second data set; After receiving the optimized second data set in the cloud, a distributed computing framework is used to perform parallel decompression and feature extraction on the second data set to obtain a third data set containing key features of device operation; Extracting fault-related features from the third data set, training and optimizing the third data set through a pre-established deep learning model to generate a fourth data set with preliminary intelligent capabilities; When the fault prediction accuracy of the fourth dataset is determined to be lower than a preset threshold in the simulation environment, the fourth dataset is iteratively updated using an incremental learning algorithm and combined with the real-time collected environmental change data to obtain a fifth dataset that can be dynamically adjusted; The improvement direction of the weak adaptive ability is obtained from the fifth data set, and the reinforcement learning algorithm is used to perform online optimization on the fifth data set to generate a sixth data set that can adapt to changes in the equipment operating environment.

2. The method for remote monitoring and fault diagnosis of automated production equipment based on the Industrial Internet according to claim 1 is characterized in that: Obtain the first data set that fully reflects the device status, including: The multi-source heterogeneous data collected by high-precision sensors during device operation is obtained by adopting multi-channel parallel processing technology to obtain an initial collection data set. The initial collection data set is then pre-processed using data fusion technology, and the multi-dimensional data is integrated using a weighted average algorithm to obtain a first data set. According to the heterogeneous data processing requirements in the first data set, determine whether the data dimension exceeds a preset threshold. If so, reduce the dimension using a principal component analysis algorithm to obtain a reduced-dimensional data set. Based on the reduced-dimensional dataset, the feature vector of the equipment's operating status is obtained. Key indicators of the status are determined through feature extraction technology. When the key indicators deviate from the normal range, the feature vector is classified through a clustering algorithm to determine the abnormal points of equipment operation. Through the mapping relationship between abnormal points and comprehensive reflection of the status, the equipment status analysis results are obtained, the status distribution characteristics are obtained, and then the time series analysis tools are used to determine the dynamic change trend of equipment operation.

3. The method for remote monitoring and fault diagnosis of automated production equipment based on the Industrial Internet according to claim 1 is characterized in that: Obtaining an optimized second data set, specifically comprising: acquiring device status information from the first data set through a sensor, extracting data using a preset sampling frequency to obtain initial state data, performing local compression on the initial state data at an edge computing node, reducing data redundancy using an LZ77 algorithm, and generating a second data set; The second data set is transmitted to the cloud via a low-latency communication protocol, and the MQTT protocol is used for data distribution. After confirming that the transmission is complete, the transmission timestamp is extracted from the second data set. When the transmission time is lower than a preset threshold, the efficiency is determined to be met, and the transmission performance index is obtained. The integrity of the data received by the cloud is analyzed based on the transmission performance indicators, and the data consistency is verified using the verification method to determine the data quality. The compression parameters of the edge computing node are adjusted according to the data quality results to generate an optimized second data set.

4. The method for remote monitoring and fault diagnosis of automated production equipment based on the Industrial Internet according to claim 1 is characterized in that: A third data set containing key characteristics of equipment operation is obtained, including: receiving an optimized second data set through the cloud, obtaining a complete data stream from a storage system, and performing sharding processing on the second data set using a distributed computing framework to obtain multiple parallel task units; Then, the parallel task units are decompressed to obtain the decompressed raw data content. The raw data content is processed using a feature extraction algorithm to obtain a feature set containing the running status. When the running status in the feature set exceeds a preset threshold, the feature set is grouped using a clustering algorithm to determine the abnormal feature subset. According to the mapping relationship between the abnormal feature subset and the equipment operation, the distribution law of the key features is determined, and then the regression algorithm is used to fit the distribution law of the key features to obtain a prediction model for the equipment operation trend.

5. The method for remote monitoring and fault diagnosis of automated production equipment based on the Industrial Internet according to claim 1 is characterized in that: Generate a fourth dataset with preliminary intelligent capabilities, including: Obtaining initial fault-related features from the third data set, processing them using feature extraction technology to obtain a first feature set, and then training them using a pre-established prediction model for equipment operation trends to obtain a preliminary training model; The first feature set is optimized through the preliminary training model to generate a second feature set. When the intelligence capability of the second feature set is lower than a preset threshold, the training process is adjusted through the optimization method to obtain a third feature set. Based on the third feature set, a fourth dataset with preliminary intelligent capabilities is constructed using data generation technology. The fourth dataset is used to verify the training effect of the deep learning model and determine whether the intelligent capabilities meet the requirements. After obtaining the verification results, the fourth data set is iteratively processed using the adjusted optimization method to obtain the final data set.

6. The method for remote monitoring and fault diagnosis of automated production equipment based on the Industrial Internet according to claim 1 is characterized in that: A fifth data set capable of dynamic adjustment is obtained, specifically including: Fault prediction is performed on the fourth data set in a simulation environment to obtain accuracy data. When the accuracy is lower than a preset threshold, the fourth data set is iteratively updated using an incremental learning algorithm to obtain an updated data set. Collect environmental change data in real time to obtain external input data. Then, through data combination methods, fuse the updated data set with the external input data to obtain a preliminary adjusted data set. Dynamic adjustment technology is used to optimize the preliminary adjusted data set to obtain the fifth data set. Fault prediction features are extracted from the fifth data set to determine whether the preset threshold requirements are met. When the fault prediction accuracy of the fifth data set is still lower than the preset threshold, the iterative update of the incremental learning algorithm is repeated to obtain the final adjusted data set.

7. The method for remote monitoring and fault diagnosis of automated production equipment based on the Industrial Internet according to claim 1 is characterized in that: Generate a sixth data set that can adapt to changes in the device operating environment, specifically including: Obtaining data extraction results from the fifth data set, using preset rules to determine features with weak adaptability, obtaining a set of improvement directions, and extracting key features from the set of improvement directions; The fifth dataset is optimized online using a reinforcement learning algorithm to generate preliminary optimized data. It then determines whether the environmental changes exceed a preset threshold. If so, the optimization parameters are adjusted to obtain data with enhanced adaptability. Based on the data with enhanced adaptability, the environmental change trend is judged and the initial version of the sixth data set is generated. Then, data extraction technology is used to separate the device operation-related features to obtain the environmental adaptation feature set. At the same time, the optimization process is applied to adjust the data structure to generate the final sixth data set.

8. The method for remote monitoring and fault diagnosis of automated production equipment based on the Industrial Internet according to claim 1 is characterized in that: Also includes: The prediction model adjustment plan is determined through the sixth data set, and the transfer learning algorithm is used to perform adaptive training on the sixth data set across device scenarios to obtain the final seventh data set.

9. The method for remote monitoring and fault diagnosis of automated production equipment based on the Industrial Internet according to claim 1, characterized in that: Also includes: Based on the seventh data set, the prediction model is deployed to the edge device through low-latency communication technology, and the device status feedback data is obtained from the edge device in real time to judge the operating stability of the remote monitoring system.

10. An industrial internet-based automated production equipment remote monitoring and fault diagnosis system, used to implement the industrial internet-based automated production equipment remote monitoring and fault diagnosis method according to any one of claims 1 to 9, characterized in that: include: The data acquisition and preprocessing module integrates multiple types of high-precision sensors to acquire multi-source heterogeneous data on equipment operation, and uses data fusion technology to preprocess the collected multi-dimensional data to obtain the first data set that fully reflects the equipment status; The data transmission and optimization module obtains device status information from the first data set, locally compresses the first data set using edge computing nodes, transmits the compressed first data set to the cloud via a low-latency communication protocol, obtains an optimized second data set, and determines the efficiency of the transmission process; A data processing and feature extraction module receives the optimized second data set from the cloud and uses a distributed computing framework to perform parallel decompression and feature extraction on the second data set to obtain a third data set containing key features of device operation; An intelligent fault diagnosis module extracts fault-related features from the third data set, trains and optimizes the third data set using a pre-established deep learning model, and generates a fourth data set with preliminary intelligent capabilities; A dynamic adjustment and optimization module, which iteratively updates the fourth dataset using an incremental learning algorithm when the fault prediction accuracy of the fourth dataset in a simulated environment is determined to be lower than a preset threshold, and combines it with the real-time collected environmental change data to obtain a fifth dataset that can be dynamically adjusted; An adaptive optimization module obtains improvement directions for weak adaptability from the fifth dataset, performs online optimization on the fifth dataset using a reinforcement learning algorithm, and generates a sixth dataset that can adapt to changes in the device operating environment; A cross-device scenario adaptation module determines a prediction model adjustment plan based on the sixth data set, and uses a transfer learning algorithm to perform adaptive training on the sixth data set for cross-device scenarios to obtain a final seventh data set; The remote monitoring and evaluation module, based on the seventh data set, deploys the prediction model to the edge device through low-latency communication technology, obtains device status feedback data from the edge device in real time, and judges the operational stability of the remote monitoring system.

Images