An AI-driven industrial IoT equipment fault prediction system

Abstract

The present invention discloses an artificial intelligence-driven industrial Internet of Things equipment fault prediction system, which relates to the field of industrial Internet of Things and predictive maintenance technology. It adopts an edge-cloud collaborative architecture to realize real-time collection, preprocessing and online fault prediction of industrial equipment sensor data. The system uses Transformer neural networks to construct a hierarchical spatiotemporal model, processes time series data in segments through a sliding window method, and integrates a causal reasoning enhancement mechanism to construct an industrial equipment causal graph to generate an attention mask, so that the model focuses on key features, thereby improving prediction accuracy and interpretability. At the same time, the system establishes a closed-loop continuous optimization mechanism, uploads edge fault prediction results and actual operation feedback to the cloud, updates the causal graph and model parameters, and realizes model adaptive optimization. The system provides efficient, accurate and explainable decision support for preventive maintenance of industrial equipment.

Description

Technical Field

[0001] The present invention relates to the field of industrial Internet of Things and predictive maintenance technology, and specifically to an artificial intelligence-driven industrial Internet of Things equipment fault prediction system. Background Art

[0002] Currently, there are several technical solutions for predictive maintenance of industrial equipment. For example, Chinese invention patent CN111401661B discloses a predictive maintenance method and maintenance system for mechanical equipment. This system collects the equipment's raw vibration acceleration and temperature data, and through preprocessing, data analysis, and fault feature extraction, predicts equipment failures and displays maintenance information. This solution effectively avoids ineffective manual work and achieves a certain degree of automated maintenance. However, it mainly relies on traditional data collection and statistical analysis methods, lacks modeling of the deep-level characteristics of the equipment's complex dynamic behavior, and has limited ability to explain the causes of failures, making it difficult to meet the requirements of high-precision prediction and detailed fault analysis.

[0003] On the other hand, Chinese invention patent CN118092404B discloses an artificial intelligence-based preventive maintenance method and system for a PLC controller network. The system deploys intelligent sensing nodes in the PLC network to collect the operating parameters of each controller in real time, and uses a preset machine learning model for preliminary analysis. It then constructs a fault prediction model based on a deep neural network to timely monitor the operating status of the equipment. However, this technology mainly relies on traditional deep learning methods for model training and data processing, and does not fully consider the mining of multi-scale spatiotemporal characteristics and causal relationships, resulting in insufficient interpretability of the prediction results. At the same time, it lacks a closed-loop adaptive optimization mechanism, making it difficult to cope with the complex and changeable actual working conditions in industrial sites.

[0004] In summary, although existing technologies have achieved predictive maintenance to a certain extent, they still have the following shortcomings: the temporal and spatial characteristics of complex equipment are not fully captured, and the prediction accuracy is difficult to meet high requirements; the prediction results lack clear fault explanations, making it difficult to accurately indicate the causes of key faults; data processing and model updates lack a closed-loop adaptive optimization mechanism, and cannot adapt to the dynamic changes of the industrial field environment in real time. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a new artificial intelligence-driven industrial Internet of Things equipment fault prediction system. By integrating Transformer hierarchical spatiotemporal modeling, causal reasoning enhancement and edge-cloud collaborative closed-loop optimization, it effectively makes up for the shortcomings of the existing technology, realizes high-precision, real-time, explainable and adaptive optimization fault prediction, and provides reliable technical support for preventive maintenance of industrial equipment.

[0006] The objective of the present invention is achieved through the following technical solutions: an artificial intelligence-driven industrial Internet of Things equipment fault prediction system, the system includes an edge device and a cloud server, and adopts an edge-cloud collaborative architecture; the edge device is used to collect sensor data of industrial equipment and preprocess it; the cloud server is used to train a fault prediction model and perform online fault prediction reasoning based on the preprocessed data; the fault prediction model uses a Transformer neural network to implement hierarchical spatiotemporal modeling, and performs segmented processing of sensor time series data through sliding window calculation; the fault prediction model integrates a causal reasoning enhancement mechanism; the causal reasoning enhancement mechanism generates an attention mask by constructing a causal graph of the operation of industrial equipment to guide the self-attention mechanism of the Transformer model to focus on key features with causal correlation; the system can monitor the operating status of industrial equipment in real time and predict the probability of fault occurrence, and provide dynamic interpretation of the prediction results based on the causal graph.

[0007] The cloud server is equipped with a causal reasoning module, which is used to build a causal graph based on the historical operating data of industrial equipment and expert knowledge. The causal graph depicts the causal relationship between the sensor parameters of industrial equipment and fault events; the causal reasoning module provides the causal graph to the Transformer fault prediction model to enhance the model's causal reasoning capabilities.

[0008] The Transformer model's self-attention mechanism applies an attention mask, which is generated based on the causal graph. By applying the attention mask to the Transformer model's self-attention calculation, the model only focuses on the relevant sensor features indicated by the causal graph at each time step, thereby improving the accuracy of fault prediction and the interpretability of the model output.

[0009] The preprocessing module of the edge device performs sliding window calculations, dividing the continuous sensor data stream into multiple time segments according to a predetermined window length and step size. Each time segment is input as a sample into the Transformer fault prediction model for analysis, and adjacent time segments overlap to ensure continuous monitoring of fault signs.

[0010] The hierarchical spatiotemporal modeling of the Transformer fault prediction model is implemented through a multi-level structure: the bottom layer is used to capture the short-term temporal patterns of each sensor data sequence; the upper layer is used to fuse the correlations and long-term trends between multiple sensors; thus, it can accurately model the complex system behavior of industrial equipment.

[0011] The system also includes a fault interpretation module, which is used to use the causal graph to determine the key sensor parameters or causal paths related to the prediction results when the equipment is predicted to have a failure risk, and output the corresponding dynamic explanation information to assist operation and maintenance personnel in analyzing the cause of the failure.

[0012] A local fault prediction unit is deployed on the edge device to run the Transformer fault prediction model trained by the cloud server, and perform real-time analysis and fault prediction on the pre-processed sensor data, thereby reducing prediction latency and reducing dependence on network connections.

[0013] The cloud server includes a model training module, which is used to collect operating data from multiple industrial IoT edge devices and perform offline training and periodic updates on the Transformer fault prediction model; the cloud server also includes a central analysis module, which is used to receive and integrate fault prediction results or alarm information from edge devices and perform centralized fault diagnosis and trend analysis.

[0014] A closed-loop continuous optimization mechanism is established between the edge device and the cloud server; the edge device sends the fault prediction results and the corresponding actual operating status feedback to the cloud server; the cloud server updates the causal graph and adjusts the parameters of the fault prediction model based on the feedback, and then sends the updated model to the edge device; thereby achieving continuous improvement and adaptive optimization of the fault prediction model performance.

[0015] The method of use includes the following steps:

[0016] S1. The edge device collects sensor data from industrial equipment and preprocesses it, dividing the sensor data into a continuous sliding window sequence;

[0017] S2. The cloud server trains the Transformer fault prediction model based on the collected historical data and builds the corresponding industrial equipment cause-and-effect graph;

[0018] S3. Deploy the trained fault prediction model and causal graph to the edge device.

[0019] S4. The edge device uses the fault prediction model to perform online fault prediction reasoning on the sensor data window obtained in real time;

[0020] S5. When the edge device detects possible fault signs or meets preset conditions, it uploads relevant sensor data and fault prediction results to the cloud server;

[0021] S6. The cloud server updates the causal graph based on the feedback data received from the edge device and retrains or adjusts the parameters of the fault prediction model;

[0022] S7. Send the updated fault prediction model to the edge device, forming a closed loop for continuous model optimization.

[0023] The beneficial effects of the present invention are:

[0024] 1. By constructing a causal graph of industrial equipment operation and generating an attention mask, the Transformer model focuses only on sensor features closely related to faults during the self-attention calculation process, effectively filtering out irrelevant information, reducing the risk of false attribution, and improving the accuracy of fault prediction. A multi-level Transformer structure is used to capture both short-term temporal changes and long-term trends in sensor data, enhancing the model's ability to express the dynamic behavior of complex industrial systems and further improving prediction performance.

[0025] 2. Edge devices independently collect and pre-process data and run local fault prediction models to achieve real-time inference of field data, ensuring a rapid response at the early stage of a fault and meeting the strict requirements of industrial sites for low-latency prediction. Through the collaborative working mode of centralized cloud training and real-time inference at the edge, it can fully utilize the high-performance computing resources of the cloud and reduce dependence on network transmission, ensuring that the system can still maintain normal operation when network conditions are unstable.

[0026] 3. Based on the constructed causal graph and attention mask, the system can output detailed dynamic explanation information for each fault prediction, clearly indicating key sensor data or causal paths, helping operations and maintenance personnel quickly understand the root cause of the fault and guide targeted maintenance. This dynamic explanation mechanism not only improves model transparency but also provides a strong basis for tracing and locating industrial equipment faults, thereby reducing the ambiguity of maintenance decisions.

[0027] 4. Edge devices feed back real-time fault prediction results and actual operating conditions to the cloud. The cloud server updates the causal graph based on the feedback data and retrains or adjusts the parameters of the Transformer model, forming a continuously optimized closed-loop system that enables the model to adapt to the ever-changing operating environment of the industrial site. Through centralized data aggregation and offline training, the cloud server can regularly update the model and causal graph to ensure that the system always adopts the latest and optimal prediction strategy, improving the long-term stability and adaptability of the overall system.

[0028] 5. By integrating fault prediction data and alarm information from multiple edge devices, cloud servers can perform centralized fault diagnosis and trend analysis for the entire industrial system, providing a macro-decision-making basis for preventive maintenance and global monitoring of large-scale industrial equipment. Global monitoring and trend analysis enable operation and maintenance personnel to identify potential failure risks in advance and take preventive measures in a timely manner, thereby reducing equipment downtime and maintenance costs, and improving production safety and efficiency.

[0029] 6. Real-time data processing and local fault prediction at the edge reduce the need for real-time data transmission to the cloud, effectively reducing the network transmission burden and improving the reliability of the system in bandwidth-constrained scenarios. The system design fully considers the resource limitations of industrial sites and the compatibility of existing PLC systems. It can be deployed without large-scale modifications, reducing implementation costs. It also has good scalability and is suitable for a variety of industrial environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 System interaction for the present invention Figure 1 ;

[0031] Figure 2 System interaction for the present invention Figure 2 . DETAILED DESCRIPTION

[0032] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0033] It is to be noted that the directions of "left", "right", "up", "down", "front", "back", "inside" and "outside" in the following schemes are all relative directions and are not listed here one by one.

[0034] Example 1:

[0035] like Figure 1 As shown, this embodiment adopts an edge-cloud collaborative architecture, which focuses on realizing real-time fault prediction functions at the edge. The cloud server is responsible for offline training and updating fault prediction models and causal graphs, while the edge devices collect data on site, perform preprocessing, local fault prediction and dynamic interpretation, thereby reducing prediction delays and reducing dependence on network connections.

[0036] Composition of edge devices

[0037] Sensor data acquisition module: Edge devices collect real-time data through various sensors installed on industrial equipment (such as temperature, pressure, vibration sensors, etc.), forming a continuous data stream.

[0038] The preprocessing module uses a sliding window calculation method to segment the continuous data stream. The specific approach is:

[0039] Preset a fixed window length (e.g. 100ms, 1 second, etc.) and overlapping step size to divide the original data into multiple time segments;

[0040] There is partial overlap between adjacent time segments, ensuring continuous capture of sudden fault signs;

[0041] Each time segment is used as an independent sample for subsequent model fault prediction analysis.

[0042] The local fault prediction unit is deployed at the edge to run the Transformer fault prediction model trained by the cloud server. After offline training, the model integrates a causal reasoning enhancement mechanism and can use the constructed industrial equipment causal graph to generate an attention mask, guiding the model to focus on key features related to the fault.

[0043] The fault prediction model uses a Transformer neural network to implement hierarchical spatiotemporal modeling. The model can encode each time segment of the input, thereby capturing local high-frequency changes and long-term trends.

[0044] The model integrates a causal reasoning mechanism: First, a causal graph of industrial equipment operation is constructed using historical data and equipment expert knowledge. Then, an attention mask is generated based on the causal graph and embedded into the Transformer self-attention layer, allowing the model to focus on sensor data with causal relevance during information interaction, thereby improving prediction accuracy and explanatory power.

[0045] Assume that the sensor data of industrial equipment is:

[0046]

[0047] where x t Represents the sensor data at time t; Constructs a causal graph G∈R n×n Its element G ij Indicates the causal influence strength of sensor i on sensor j, for example, G ij =0.8G means that the change of sensor i has a strong causal effect on j.

[0048] For the sliding window W (k) For data within ⊂X (window length Δ), the query matrix Q, key matrix K, and value matrix V are calculated through the Transformer model.

[0049] W (k) The data subsequence (time segment) within the kth sliding window is usually of length Δ, and windows can overlap. It is used to locally model time series features.

[0050] The length of the Δ sliding window, that is, how many time steps each subsequence contains (such as 100 means 0.5 seconds).

[0051] k represents the number of the kth sliding window, ranging from 1 to T−Δ+1, which is used to traverse the entire time series.

[0052] G ij Edge weights in a causal graph. Larger values ​​indicate a stronger influence of variable i on variable j. They can be positive (promoting causation), negative (inhibitory causation), or 0 (no direct causal relationship).

[0053] The causal fusion operator ⊕ is introduced and defined as:

[0054]

[0055] Used to fuse traditional dot-product attention scores with causal influence information. This operation provides normalization and co-regulation, enhancing the nonlinear expression of the synergistic effect between two variables. |ϵ| is a small constant used to prevent the denominator from being zero or too small, typically set to 1e−6.

[0056] Design a causal conversion function Γ to transform G ij Perform nonlinear compression and smoothing to adapt it to the range of attention weights and enhance the sensitivity of medium-intensity causal information. The function formula is as follows:

[0057]

[0058] It integrates positive and negative causal information and has strong differentiability and smoothness.

[0059] Improved self-attention weights The attention weight of position i on position j at time t combines the attention score and causal information. It is used to measure which sensor channel features the Transformer should focus on at a specific time position. It is defined as follows:

[0060]

[0061] Compared with traditional Transformer, causal weights are directly involved in the normalized softmax weighting mechanism. The output representation of the model at time step t and channel i is a linear combination of all attention-weighted Value vectors, namely:

[0062]

[0063] T is the total number of time steps, which represents the length of the collected sensor data sequence. For example, if the data is collected continuously for 10 seconds and sampled 10 times per second, then T=100T = 100T=100.

[0064] t is the current time step, 1≤t≤T, which represents a specific time point in the entire time series data.

[0065] x t ∈Rn is the data vector collected by n sensors at time t, where each component is the observation value of the i-th sensor.

[0066] The entire sequence of raw sensor data, of length T, is the main input of the model.

[0067] n is the total number of sensors deployed on industrial equipment, that is, the data dimension at each point in time. Common channels include temperature, pressure, current, voltage, speed, vibration, and so on.

[0068] ∈R d is the query vector at the tth time step and position i in the Transformer, which is input Obtained through linear transformation or embedding module, the dimension is d. It is used to initiate attention query for information at other locations.

[0069] ∈Rd is the key vector of the t-th time step and position j in the Transformer, which is used to calculate the degree of attention matching with the query vector.

[0070] ∈Rd is the value vector of the t-th time step and position j, which is weighted by the attention weight and participates in the final representation output.

[0071] d is the dimension of the vector, which is usually the latent space dimension set by the model (such as 64, 128, etc.), used to control the representation capability and computational scale.

[0072] i, j, k represent different feature channel indices, ranging from 1≤i, j, k≤n, and are used for cross-channel modeling in attention calculation.

[0073] Formula overall structure process

[0074] Input data: multi-channel time series sensor data X={xt}

[0075] Sliding window processing: segmented extraction of local time slices W (k)

[0076] Transformer encoding: extract the Q, K, and V vectors for each segment of data

[0077] Introducing causal graph G: Constructed based on experience or data deduction

[0078] Causally Enhanced Attention Computation:

[0079] Dot product attention score Q T K√d

[0080] Nonlinear transformation causal weight Γ(G ij )

[0081] Fusion Operations

[0082] Softmax normalization results

[0083] Output representation :Serves as input for prediction and explanation

[0084] Working process

[0085] Data collection and preprocessing

[0086] Edge devices collect sensor data from industrial equipment in real time;

[0087] The preprocessing module uses a sliding window method to divide the continuous data stream into multiple overlapping time segments according to a predetermined window length and step size, and each time segment constitutes a sample.

[0088] Local model inference

[0089] Each time segment sample is input into the Transformer fault prediction model deployed at the edge;

[0090] The model uses the attention mask generated by the pre-embedded causal graph to guide the self-attention mechanism to focus on key sensor features and extract local spatiotemporal patterns closely related to faults.

[0091] The model outputs real-time failure probability and provides key feature information related to the prediction results through the dynamic interpretation function of the attention mechanism.

[0092] Fault prediction and response

[0093] The edge fault prediction unit triggers local alarms or control instructions based on the fault probability output by the model, supporting real-time monitoring of industrial equipment and emergency response to faults.

[0094] At the same time, prediction results and dynamic interpretation information can be uploaded to the cloud server when network conditions permit, to facilitate centralized management, historical data analysis and continuous model optimization.

[0095] By deploying local fault prediction units at the edge, instant analysis and fault diagnosis of segmented samples can be achieved, significantly reducing data transmission and processing delays and meeting the low-latency response requirements of industrial sites.

[0096] The edge independently performs fault prediction to avoid prediction interruptions due to network delays or instability, ensuring continuous monitoring and safe operation of field equipment even in environments with poor network connectivity.

[0097] After integrating the causal reasoning enhancement mechanism, the Transformer model uses the attention mask generated by the causal graph to enable the model to focus only on features with actual physical and causal correlations, thereby improving the accuracy of fault prediction; at the same time, the dynamic interpretation function helps operation and maintenance personnel quickly understand the basis for the prediction, making it easier to take targeted maintenance measures in a timely manner.

[0098] The data stream is segmented using sliding window calculations to allow adjacent time segments to overlap, ensuring continuous capture of device status changes, timely detection of minor anomalies, and early warning of fault trends.

[0099] In general, this embodiment effectively achieves low-latency prediction and rapid on-site response to industrial IoT equipment failures by implementing sensor data preprocessing, real-time model reasoning, and dynamic interpretation at the edge, providing reliable technical support for the safe and stable operation of industrial equipment.

[0100] Example 2:

[0101] like Figure 2 As shown, based on Example 1, the entire system of this embodiment still adopts an edge-cloud collaborative architecture, but the focus is on centralized processing of the cloud server. The cloud server integrates a causal reasoning module, a model training module, a central analysis module, and a fault interpretation module, and uses large-scale historical data and multi-edge device feedback to achieve offline training, periodic updates, and causal reasoning enhancement of the Transformer fault prediction model.

[0102] Cloud server composition

[0103] The causal reasoning module uses historical operation data and expert knowledge to build a causal graph of industrial equipment operation.

[0104] The causal graph depicts the causal relationship between each sensor parameter and the fault event, and generates corresponding attention masks for use by the Transformer model to enhance the model's causal reasoning capabilities.

[0105] The model training module aggregates historical operating data from multiple edge devices and performs offline training of the Transformer fault prediction model.

[0106] The model adopts a hierarchical spatiotemporal modeling structure:

[0107] The bottom layer captures the temporal patterns of short-term sensor data sequences;

[0108] The upper layer fuses correlations and long-term trends among multiple sensors to accurately model the complex behavior of industrial equipment.

[0109] During training, the attention mask generated by the causal graph is embedded into the self-attention layer of the Transformer to ensure that the model only focuses on the relevant features indicated by the causal graph at each time step.

[0110] The fault interpretation module uses the pre-built causal graph to determine the key sensor parameters or causal paths related to the prediction results when the model predicts that the equipment has a failure risk, and generates dynamic explanation information to assist operation and maintenance personnel in analyzing the cause of the failure.

[0111] The central analysis module aggregates and integrates fault prediction results and alarm information from multiple edge devices, performs centralized fault diagnosis and trend analysis, and provides macro-level fault situation monitoring and prediction support for the entire industrial system.

[0112] Edge and cloud working together

[0113] Data upload and feedback: The edge device uploads the pre-processed data and real-time fault prediction results to the cloud.

[0114] The cloud server continuously updates the causal graph and Transformer model based on feedback data from each edge device, achieving periodic retraining and performance optimization of the model.

[0115] Model distribution and update: The latest Transformer fault prediction model trained by the cloud model training module will be distributed to the edge device to ensure that the on-site fault prediction module always uses the latest and optimized model.

[0116] Working process

[0117] Causal graph construction and attention mask generation

[0118] The cloud server first uses massive historical data and expert input to build a causal graph of industrial equipment through a causal reasoning module.

[0119] Based on the constructed causal graph, the system generates a corresponding attention mask and provides it to the Transformer model, so that the model only focuses on sensor features related to fault prediction during the self-attention calculation process.

[0120] Offline model training and updating

[0121] The model training module collects historical data from multiple edge devices and uses a multi-level Transformer model to perform offline training on the data.

[0122] During training, by utilizing attention masks and a hierarchical spatiotemporal modeling structure, the model is able to accurately capture both short-term signal fluctuations and long-term trends.

[0123] After training is completed, the cloud server evaluates and verifies the model, and periodically retrains and updates the model.

[0124] Fault Interpretation and Central Analysis

[0125] When the cloud model detects a potential failure risk in the device, the fault explanation module uses the causal graph to explain the model prediction results and generate a dynamic explanation report of key sensor parameters or causal paths.

[0126] At the same time, the central analysis module integrates fault prediction and alarm information from each edge device, conducts centralized fault diagnosis and trend analysis, and forms a global fault monitoring report to provide data support for decision-making.

[0127] Model delivery and edge collaboration

[0128] The latest trained model is sent to the edge device through a secure network channel.

[0129] Edge devices use the latest models to perform real-time fault prediction and feed the results back to the cloud, forming a closed loop for continuous optimization of data and models.

[0130] Through centralized model training in the cloud and the use of massive historical data and feedback from multiple edge devices, the Transformer model's ability to model the behavior of complex industrial equipment can be greatly improved; at the same time, the causal reasoning enhancement mechanism ensures that the model only focuses on truly relevant key features in fault prediction, thereby improving prediction accuracy.

[0131] By integrating causal graphs and attention masking mechanisms, the fault explanation module can output detailed dynamic explanation information for each prediction, helping operation and maintenance personnel quickly locate the root cause of the fault, enhancing the system's interpretability and operational confidence.

[0132] The cloud-based central analysis module aggregates data from multiple edge devices, enabling fault diagnosis and trend analysis of the entire industrial system, providing a macro-decision-making basis for equipment maintenance and preventive repairs, and reducing overall operation and maintenance risks.

[0133] Based on the continuously collected feedback data from edge devices, the cloud server can periodically retrain the causal graph and Transformer model to achieve model adaptive optimization and continuously improve the overall performance and stability of the system.

[0134] Through centralized fault diagnosis and trend analysis in the cloud, operations and maintenance personnel can take targeted maintenance measures more promptly, detect potential faults in advance, and perform preventive repairs, thereby significantly reducing equipment downtime and maintenance costs.

[0135] In general, this embodiment achieves high-precision, global management and dynamic response to fault prediction of industrial IoT devices by centrally building causal graphs, offline training Transformer models and dynamic fault interpretation in the cloud. The cloud and edge work together to not only improve the accuracy and explanation capabilities of fault prediction, but also provide strong guarantees for the safe and stable operation of large-scale industrial systems.

[0136] Example 3:

[0137] like Figure 1 and Figure 2 As shown, this embodiment still adopts the edge-cloud collaborative architecture based on Examples 1 and 2, but focuses on centralized processing and closed-loop optimization of cloud servers. The entire system consists of edge devices and cloud servers. The edge is mainly responsible for data collection, preprocessing and real-time online fault prediction, while the cloud server is responsible for offline model training, causal graph construction, fault interpretation and model adaptive optimization. The main components include:

[0138] The cloud-based causal reasoning module uses historical operating data of industrial equipment and expert knowledge to construct a causal graph, characterizing the causal relationship between each sensor parameter and fault event, and generating a corresponding attention mask. This attention mask is applied in the self-attention layer of the Transformer model, allowing the model to focus only on the relevant features indicated by the causal graph at each time step, thereby improving the accuracy and interpretability of fault prediction.

[0139] The cloud-based model training module aggregates historical data from multiple edge devices and uses a Transformer neural network to implement hierarchical spatiotemporal modeling, including:

[0140] The bottom layer is used to capture short-term temporal patterns in sensor data sequences;

[0141] The upper layer is used to fuse the correlation and long-term trends among multiple sensors.

[0142] During the training process, the causal reasoning enhancement mechanism is embedded in the model to optimize the model parameters and ensure that the model has high predictive performance.

[0143] Closed-loop continuous optimization mechanism: edge devices upload real-time fault prediction results and actual equipment operating status feedback to the cloud server. The cloud server uses this feedback data to update the causal graph and retrain or adjust the parameters of the Transformer model. The updated model is sent to the edge device through a secure channel to achieve continuous improvement and adaptive optimization of model performance.

[0144] The central analysis module integrates fault prediction data and alarm information from multiple edge devices, performs centralized fault diagnosis and trend analysis on the entire industrial system, and provides macro support for system optimization and decision-making.

[0145] In addition, this embodiment uses the following method steps to achieve closed-loop operation of the entire system:

[0146] S1: Data acquisition and preprocessing

[0147] Edge devices collect sensor data from industrial equipment and divide the data into continuous sliding window sequences to ensure that data samples have temporal continuity and overlap.

[0148] S2: Offline model training and causal graph construction

[0149] The cloud server uses historical data to train the Transformer fault prediction model and constructs the corresponding causal graph of industrial equipment to provide causal prior information for model training.

[0150] S3: Model and Causal Graph Deployment

[0151] After training is completed, the cloud server will send the updated fault prediction model and causal graph to the edge device to ensure that the on-site fault prediction module always uses the latest optimized model.

[0152] S4: Online Fault Prediction Reasoning

[0153] The edge device uses the distributed model to perform online fault prediction on the real-time collected data window and outputs the fault probability and dynamic interpretation results.

[0154] S5: Upload abnormal feedback

[0155] When the edge device detects a potential fault or reaches the preset alarm threshold, the relevant sensor data and prediction results are uploaded to the cloud server in a timely manner.

[0156] S6: Cloud-based feedback processing and model updating

[0157] Based on the feedback data received from the edge, the cloud server updates the causal graph and retrains or adjusts the parameters of the Transformer model to adapt to the device operating status and environmental changes.

[0158] S7: Model distribution and continuous optimization

[0159] The updated model is sent from the cloud server to the edge device, forming a closed loop of continuous optimization to ensure continuous improvement of the overall prediction performance of the system.

[0160] Working process

[0161] Initial model training and causal graph construction

[0162] The cloud server first uses historical data and expert knowledge to build an initial causal graph and generates the corresponding attention mask.

[0163] At the same time, the model training module uses the Transformer neural network to perform offline training on historical data to obtain the initial fault prediction model.

[0164] After training is completed, the cloud deploys the fault prediction model and causal graph to the edge device (steps S2 and S3).

[0165] Real-time online fault prediction

[0166] The edge device collects real-time sensor data and divides it into data samples according to the sliding window through the preprocessing module (step S1).

[0167] Each sample runs the Transformer model sent by the online fault prediction unit at the edge to perform fault prediction reasoning and output the fault probability and corresponding dynamic explanation (step S4).

[0168] Anomaly detection and feedback upload

[0169] When the edge device detects that the prediction result reaches or exceeds the preset fault threshold, the edge device uploads the relevant data and prediction results to the cloud server (step S5).

[0170] Closed-loop optimization and model updating

[0171] The feedback processing module of the cloud server receives abnormal data and actual operating conditions from the edge devices, updates the causal graph, and uses this data to retrain or fine-tune the parameters of the fault prediction model (step S6).

[0172] The updated model is sent to the edge device through a secure channel, replacing the old model and achieving closed-loop continuous optimization (step S7).

[0173] Central analysis and global monitoring

[0174] At the same time, the central analysis module integrates data from multiple edge devices, performs global fault diagnosis and trend analysis, and forms a comprehensive report to provide support for operation and maintenance decisions.

[0175] Through centralized cloud-based training, model optimization using large amounts of historical data and feedback from multiple edge devices, and the embedding of causal reasoning enhancement mechanisms, the Transformer model can more accurately capture the complex spatiotemporal behavior of industrial equipment and significantly improve fault prediction accuracy.

[0176] By generating an attention mask based on the constructed causal graph, the model focuses on key sensor features related to the fault during reasoning, so that each prediction can output dynamic explanatory information, helping operation and maintenance personnel to identify the cause of the fault and key influencing factors, thereby improving the overall interpretability of the system.

[0177] The closed-loop continuous optimization mechanism established between edge devices and cloud servers enables the system to timely adjust the causal graph and model parameters based on actual equipment operation feedback, achieve adaptive model improvement, and effectively respond to the ever-changing working conditions and data distribution in the industrial environment.

[0178] The cloud server integrates the operating data of multiple edge devices to realize centralized fault diagnosis and trend analysis of the entire industrial system, provide macro-decision-making support for preventive maintenance and equipment management, and reduce operation and maintenance risks and downtime costs.

[0179] By utilizing the edge-cloud collaborative architecture, we can fully leverage the advantages of large-scale computing and storage in the cloud, and achieve real-time reasoning at the edge, ensuring that the system can still operate normally when network conditions are unstable, while saving transmission bandwidth and on-site processing resources.

[0180] In general, this embodiment achieves high-precision prediction and global monitoring of industrial equipment failures through centralized cloud-based causal graph construction, offline training of Transformer models, dynamic fault interpretation, and closed-loop continuous optimization. The cloud and edge work together to not only improve the system's prediction performance and explainability, but also achieve adaptive optimization and centralized management, providing strong technical support for the stable operation of industrial IoT equipment.

[0181] The above description is only a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein and should not be regarded as excluding other embodiments. Instead, it can be used in various other combinations, modifications and environments, and can be modified within the scope of the concept described herein through the above teachings or technology or knowledge in related fields. The changes and modifications made by those skilled in the art do not depart from the spirit and scope of the present invention and should be within the scope of protection of the claims attached to the present invention.

Claims

1. An artificial intelligence-driven industrial Internet of Things equipment fault prediction system, characterized by: The system includes edge devices and cloud servers, and adopts an edge-cloud collaborative architecture; The edge device is used to collect sensor data of industrial equipment and preprocess it; the cloud server is used to train the fault prediction model and perform online fault prediction reasoning based on the preprocessed data; the fault prediction model uses the Transformer neural network to implement hierarchical spatiotemporal modeling, and performs segmented processing of sensor time series data through sliding window calculation; the fault prediction model integrates a causal reasoning enhancement mechanism; the causal reasoning enhancement mechanism generates an attention mask by constructing a causal graph of industrial equipment operation to guide the self-attention mechanism of the Transformer model to focus on key features with causal correlation, and constructs a causal graph of industrial equipment operation through historical data and equipment expert knowledge; then, an attention mask is generated based on the causal graph and embedded in the Transformer self-attention layer, so that the model focuses on sensor data with causal correlation during information interaction, thereby improving prediction accuracy and explanation ability; the system can monitor the operating status of industrial equipment in real time and predict the probability of fault occurrence, and provide dynamic explanations for the prediction results based on the causal graph.

2. The artificial intelligence-driven industrial IoT equipment fault prediction system according to claim 1, characterized in that: The cloud server is equipped with a causal reasoning module for constructing a causal graph based on the historical operating data of the industrial equipment and expert knowledge. The causal graph depicts the causal relationship between various sensor parameters of the industrial equipment and fault events; The causal reasoning module provides the causal graph to the Transformer fault prediction model to enhance the causal reasoning capability of the model.

3. The artificial intelligence-driven industrial IoT equipment fault prediction system according to claim 2, characterized in that: An attention mask is applied to the self-attention mechanism of the Transformer model, and the attention mask is generated based on the causal graph. By applying the attention mask to the self-attention calculation of the Transformer model, the model only focuses on the relevant sensor features indicated by the causal graph at each time step, thereby improving the accuracy of fault prediction and the interpretability of the model output.

4. The artificial intelligence-driven industrial Internet of Things equipment fault prediction system according to claim 3, characterized in that: The preprocessing module of the edge device performs sliding window calculations to divide the continuous sensor data stream into multiple time segments according to a predetermined window length and step size; each time segment is input into the Transformer fault prediction model as a sample for analysis, and adjacent time segments overlap to ensure continuous monitoring of fault signs.

5. The artificial intelligence-driven industrial Internet of Things equipment fault prediction system according to claim 2, characterized in that: The hierarchical spatiotemporal modeling of the Transformer fault prediction model is implemented through a multi-level structure: the bottom layer is used to capture the short-term temporal patterns of each sensor data sequence; the upper layer is used to fuse the correlations and long-term trends between multiple sensors, thereby enabling accurate modeling of the complex system behavior of industrial equipment.

6. The artificial intelligence-driven industrial Internet of Things equipment fault prediction system according to claim 5, characterized in that: The system also includes a fault interpretation module for, when predicting a failure risk of the equipment, using the causal graph to determine key sensor parameters or causal paths related to the prediction result, and outputting corresponding dynamic interpretation information to assist operation and maintenance personnel in analyzing the cause of the failure.

7. The artificial intelligence-driven industrial Internet of Things equipment fault prediction system according to claim 6, characterized in that: A local fault prediction unit is deployed on the edge device, which is used to run the Transformer fault prediction model trained by the cloud server, perform real-time analysis and fault prediction on the preprocessed sensor data, thereby reducing prediction delay and reducing dependence on network connection.

8. The artificial intelligence-driven industrial Internet of Things equipment fault prediction system according to claim 7, characterized in that: The cloud server includes a model training module for collecting operating data from multiple industrial Internet of Things edge devices and performing offline training and periodic updates on the Transformer fault prediction model; the cloud server also includes a central analysis module for receiving and integrating fault prediction results or alarm information from edge devices and performing centralized fault diagnosis and trend analysis.

9. The artificial intelligence-driven industrial Internet of Things equipment fault prediction system according to claim 7, characterized in that: A closed-loop continuous optimization mechanism is established between the edge device and the cloud server; the edge device sends fault prediction results and corresponding actual operating status feedback to the cloud server; The cloud server updates the causal graph based on the feedback and adjusts the parameters of the fault prediction model, and then sends the updated model to the edge device; thereby achieving continuous improvement and adaptive optimization of the performance of the fault prediction model.

10. The artificial intelligence-driven industrial Internet of Things equipment fault prediction system according to claim 1, characterized in that: The method of use includes the following steps: S1. The edge device collects sensor data of industrial equipment and preprocesses it, dividing the sensor data into a continuous sliding window sequence; S2. The cloud server trains the Transformer fault prediction model based on the collected historical data and builds the corresponding industrial equipment cause-and-effect graph; S3. Deploy the trained fault prediction model and the causal graph to the edge device; S4. The edge device uses the fault prediction model to perform online fault prediction reasoning on the sensor data window obtained in real time; S5. When the edge device detects possible fault signs or meets preset conditions, it uploads relevant sensor data and fault prediction results to the cloud server; S6. The cloud server updates the causal graph based on the feedback data received from the edge device, and retrains or adjusts parameters of the fault prediction model; S7. Send the updated fault prediction model to the edge device, forming a closed loop for continuous model optimization.

Images