Intelligent operation and maintenance platform based on deep learning and industrial Internet of Things

Abstract

The invention discloses an intelligent operation and maintenance platform based on deep learning and an industrial internet of things, and belongs to the field of intelligent industrial control. The platform adopts a layered and distributed architecture and comprises an equipment layer, an acquisition and transmission layer, a data communication layer and a service application layer, the equipment layer acquires equipment operation data through a multi-source sensor; the acquisition and transmission layer is responsible for data primary processing and transmission; the data communication layer integrates, standardizes and distributes heterogeneous data; the business application layer provides an operation and maintenance management function, the core of the business application layer is to integrate an expert system module, and the expert system module performs time domain, frequency domain and time-frequency domain feature extraction on equipment multi-source signals and inputs the time domain, frequency domain and time-frequency domain features into a trained multi-source parallel convolution-time sequence dependence capture-attention mechanism weighted fusion deep learning model to obtain a multi-source parallel convolution-time sequence dependence capture-attention mechanism weighted fusion deep learning model. And intelligent diagnosis of equipment faults and accurate prediction of residual life are realized. Predictive maintenance and intelligent operation and maintenance of the industrial equipment are realized, and the operation and maintenance efficiency and the equipment reliability are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent industrial control, and in particular to a smart operation and maintenance platform based on deep learning and the Industrial Internet of Things. Background Technology

[0002] With the increasing maturity of IoT technology, industrial production is accelerating its transformation towards intelligence. By integrating various sensors and controllers with sensing and control capabilities into the production process, the Industrial Internet of Things (IIoT) has significantly improved the level of process supervision and demonstrated great potential in reducing accident risks and labor costs.

[0003] However, most current industrial site monitoring and control systems still rely on traditional hardware architectures, with their functions largely limited to remote monitoring and basic management. Maintenance work remains highly dependent on manual experience, contradicting the principles of intelligent and automated maintenance. This model is inefficient, particularly lacking timely diagnosis, trend prediction, and early warning mechanisms for equipment anomalies, leading to frequent unexpected equipment failures and unplanned downtime. Summary of the Invention

[0004] To address the above problems, this invention provides a smart operation and maintenance platform based on deep learning and the Industrial Internet of Things. Through a layered distributed architecture and a built-in intelligent expert system, it achieves deep insight into equipment status and predictive maintenance, aiming to fundamentally solve the problems of low operation and maintenance efficiency and unplanned downtime caused by the reliance on manual labor and lack of early warning capabilities in existing systems.

[0005] This invention provides a smart operation and maintenance platform based on deep learning and the Industrial Internet of Things, comprising: The equipment layer is used to collect operational data of production equipment through a variety of sensors and actuators integrated in the industrial field; The acquisition and transmission layer is used for preliminary processing of the acquired data; The data communication layer is used to integrate and standardize the data after preliminary processing. The business application layer provides operation and maintenance management functions based on the data processed by the data communication layer. It includes an expert system module, which uses a deep learning model to assess the health status of the equipment based on the equipment operation data.

[0006] Furthermore, the workflow of the expert system module is as follows: Acquire heterogeneous data after it has been processed by the data communication layer; Heterogeneous data is preprocessed and features are extracted, including time-domain features, frequency-domain features, and time-frequency-domain features. The extracted features are input into a trained deep learning model for equipment fault diagnosis or lifespan prediction.

[0007] By clearly defining the workflow of expert system data acquisition, preprocessing and feature extraction, and model inference, and by comprehensively extracting time-domain, frequency-domain, and time-frequency-domain features, the limitations of single-signal or single-dimensional features are overcome. This enables a comprehensive characterization of equipment health status from multiple dimensions, significantly improving the accuracy of fault diagnosis and the reliability of lifespan prediction.

[0008] Furthermore, the deep learning model is a neural network model that combines multi-source parallel convolution, temporal dependency capture, and attention mechanisms in a weighted manner.

[0009] This study defines a deep learning model that employs a weighted fusion structure of multi-source parallel convolution, temporal dependency capture, and attention mechanisms. This model can process multi-source signals in parallel, fully leveraging the advantages of deep learning in automatically mining deep features and effectively capturing long-term temporal patterns and focusing on key information. This enables accurate identification of complex failure modes and robust prediction of performance degradation trends.

[0010] Furthermore, the training process of the deep learning model includes: The preprocessed heterogeneous data is divided into training set, validation set and test set, and the neural network model is constructed accordingly. The model was trained using the Adam optimizer, and an early stopping strategy was employed to prevent overfitting. The model is tuned using the validation set and its performance is evaluated using the test set.

[0011] By employing a standardized model training process, including dataset partitioning, using the Adam optimizer, and adopting an early stopping strategy, the scientific rigor and stability of the model training process are ensured. This method effectively prevents model overfitting, improves its generalization ability, and thus guarantees that the final deployed model maintains high accuracy and reliability when facing unknown data.

[0012] Furthermore, the deep learning model includes: Multi-source independent convolutional layers are used to process signals from different sources using convolutional kernels of different sizes, in order to extract local features of different signals; Pooling layers are used to reduce the dimensionality of the outputs of multi-source independent convolutional layers to condense key features; A bidirectional LSTM layer is used to capture long-term temporal dependencies of device states and output temporal features. An attention mechanism layer is used to weight the temporal features to focus on features at key moments.

[0013] By constructing a dedicated neural network consisting of multi-source independent convolutional layers, pooling layers, bidirectional LSTM layers, and attention mechanism layers, this model can specifically process industrial time-series data. It can extract the local spatiotemporal features of different signals, understand the long-term state evolution patterns, and highlight key moments through the attention mechanism, ultimately achieving a more refined and accurate assessment of equipment failure and remaining life.

[0014] Furthermore, the process of using convolutional kernels of different sizes to process signals from different sources specifically involves: Small convolutional kernels are used to capture high-frequency impact features caused by faults, while large convolutional kernels are used to capture low-frequency trend features of lifetime degradation.

[0015] By specifically defining the role of convolutional kernels of different sizes, the design of the neural network is combined with the physical mechanism of equipment failure. This allows the model to simultaneously and sensitively respond to subtle signs of sudden failures, while also robustly grasping the slow process of long-term performance degradation, thus enhancing the interpretability and engineering applicability of the model.

[0016] Furthermore, the business application layer also includes: The equipment management module is used for equipment information entry, historical equipment data query, and production process management. The alarm management module is used to receive abnormal information output by the expert system module and issue fault alarms, while also providing query functions for fault information and platform historical events. The report management module is used to automatically generate energy consumption reports and equipment utilization reports based on the conditions selected by the user. The system management module is used to manage the basic information of platform users and configure access and operation permissions for different accounts.

[0017] By integrating modules for device management, alarm management, report management, and system management, a fully functional operation and maintenance management ecosystem has been built, improving the integration and efficiency of operation and maintenance management.

[0018] Furthermore, the data communication layer includes: The core bus is used for aggregating, standardizing, and distributing heterogeneous data in real time. Distributed data communication systems are deployed in factory workshops or at the edge of production lines to collect and preprocess data locally.

[0019] Through the collaborative design of the core bus and the distributed data communication system, a "center-edge" collaborative data processing system was constructed. This system not only ensures the unified management and efficient distribution of global data, but also effectively reduces network bandwidth pressure and data transmission latency through local preprocessing at the edge, improves the system's real-time response capability to local abnormal events, and enhances the system's robustness during network fluctuations.

[0020] Compared with existing technologies, the beneficial effects of this invention are as follows: By constructing a collaborative architecture comprising a device layer, a data acquisition and transmission layer, a data communication layer, and a business application layer, a full-link intelligent operation and maintenance system integrating data perception, edge processing, intelligent analysis, and decision response is formed. This platform not only achieves plug-and-play functionality and unified data integration for multi-source heterogeneous devices, but also leverages multi-dimensional feature extraction from expert systems and dedicated deep learning models to deeply integrate time-domain, frequency-domain, and time-frequency-domain features. Furthermore, it employs multi-source convolution, bidirectional LSTM, and attention mechanisms to accurately capture local anomalies and long-term evolution patterns of device status, thus demonstrating superior accuracy and robustness in early fault diagnosis and remaining lifespan prediction. Based on this, the system seamlessly integrates intelligent analysis results with business modules such as equipment management, real-time alarms, and energy efficiency reports, driving the operation and maintenance model from traditional "post-event handling" to "predictive maintenance." This significantly reduces unplanned downtime, lowers the safety risks and maintenance costs associated with sudden failures, and reduces manpower input through automated reporting and refined management, thereby improving overall operation and maintenance efficiency and productivity. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this drawing or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this drawing. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0022] Figure 1 This is an architecture diagram of the intelligent operation and maintenance platform based on deep learning and industrial IoT of the present invention; Figure 2 This is a flowchart of the expert system module of the present invention; Figure 3 This is a diagram showing the interaction relationships between the modules of the business application layer of this invention; Figure 4 This is a flowchart illustrating the training process of the deep learning model of the present invention. Figure 5 This is a network structure diagram of the deep learning model of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments provided by this invention without inventive effort are within the scope of protection of this invention.

[0024] This invention provides a smart operation and maintenance platform based on deep learning and the Industrial Internet of Things, such as... Figure 1 As shown, the platform's technical architecture, from bottom to top, includes a device layer, a data acquisition and transmission layer, a data communication layer, and a business application layer, forming a complete technology chain of "device perception - data acquisition - network transmission - intelligent decision-making." Through a layered and decoupled design, it enables plug-and-play functionality for heterogeneous devices and supports the rapid deployment of new device types using a Model-as-a-Service (MaaS) model, forming a scalable industrial intelligent operation and maintenance solution.

[0025] The device layer is used to collect operational data of production equipment through a variety of sensors and actuators integrated in the industrial field.

[0026] Specifically, the equipment layer, as the system foundation, integrates various sensors (temperature, vibration, pressure, etc.), actuators, and industrial equipment in the industrial field. It is used to collect and display various operational data of the equipment during the production process, including multi-source data such as motor vibration data, temperature data, and voltage and current. Furthermore, it enables real-time monitoring of equipment status and command feedback through standardized interfaces (such as Modbus / OPC UA). Its modular design facilitates the access of heterogeneous devices and protocol conversion.

[0027] The acquisition and transmission layer is used to perform preliminary processing on the acquired data.

[0028] Specifically, the data acquisition and transmission layer plays a crucial role in the platform, primarily responsible for the initial processing of data from the device layer and its secure and reliable transmission to the upper-layer platform. This layer is built upon the actual needs of industrial sites and incorporates design principles of traditional Industrial IoT platforms, typically including core functions such as data acquisition interfaces, protocol conversion, data caching, and transmission. Furthermore, the acquisition and transmission layer also possesses certain edge processing capabilities, enabling it to perform preliminary judgment and filtering of the acquired data. For example, it can determine whether abnormal data needs to be reported immediately based on the device's operating status, thereby reducing invalid data traffic and improving overall system efficiency.

[0029] The data communication layer is used to integrate and standardize the data after preliminary processing.

[0030] Specifically, the data communication layer comprises a core bus and a distributed data communication system. The core bus serves as the central hub of the data communication layer, its core function being to aggregate, standardize, and efficiently transmit heterogeneous data to support the real-time data needs of upper-layer applications, such as equipment monitoring, predictive maintenance, and AI-driven quality inspection. The distributed data communication system is the edge extension of the data communication layer, deployable in factory workshops, production lines, and even near equipment. Its core function is to collect data locally and preprocess it locally, reducing the latency and bandwidth pressure of centralized systems. The combination of these two systems allows for flexible expansion, constructing a seamless data flow network from the "device layer to the platform layer," providing high-quality, real-time data support for the "intelligent decision-making" of industrial IoT platforms.

[0031] The business application layer is used to provide operation and maintenance management functions based on the data processed by the data communication layer. It includes an expert system module, which uses a deep learning model to assess the health status of the equipment based on the equipment operation data.

[0032] Specifically, the business application layer is the platform's capability hub, serving as its "user interface" and "value output terminal." It directly addresses the actual needs of operations and maintenance (O&M) personnel and managers, transforming the equipment data collected at the lower levels into actionable O&M decisions. Functionally, the business application layer can be categorized into three main types: auxiliary management, real-time monitoring, and expert systems.

[0033] The auxiliary management category includes a series of functional management modules such as equipment management, alarm management, report management, and system management, such as... Figure 3 As shown, the main responsibilities include the operation and maintenance management of the platform and users, ensuring the normal operation and management of the platform, and providing solutions for various user needs.

[0034] The equipment management module is used for equipment information entry, historical data query, and production process management. Equipment information entry involves inputting basic equipment model information, various functional parameters, and related accessory information. This information is displayed in a cell list format combining text and graphics. Users can click on equipment details to view the equipment's real-time status. The historical data query provides various comparative displays of historical equipment operating data. Users can select equipment operating parameters as needed and save the corresponding historical data locally. Furthermore, users can perform conditional searches using forms based on equipment number, group, time, maximum and minimum values, and energy consumption. Historical data is displayed in tables, pie charts, bar charts, and donut charts, and the corresponding charts can be downloaded and exported. The production resource configuration function is responsible for the rational allocation and adjustment of various resources during industrial production, and for assigning appropriate personnel to the production process to ensure smooth operation. Typically, production process management is displayed using a production line diagram, based on the equipment functional parameters and resource usage at each node.

[0035] The alarm management module is used for fault alarms, fault queries, and event queries. Fault alarms primarily rely on the output of the expert system module. Once an anomaly is detected, an alarm will be issued immediately and relevant information will be recorded. Alarm information will be displayed in the form of pop-ups, voice prompts, and alarm animations for timely user access. Fault queries mainly rely on historical alarm information, allowing users to search by time, device, and group number. This includes both operational alarm queries and fault alarm queries. Users can perform fuzzy searches and conditional searches, and the results are displayed in a table format, with detailed information for each result. Event queries provide the ability to query historical event information on the platform. Users can search by time, device, and group number, primarily involving login queries and maintenance queries. Users can perform fuzzy searches and conditional searches, and the results are displayed in a table format.

[0036] The report management module is used to generate energy consumption reports and utilization rate reports. Users can select conditions such as date and equipment; the backend calculates energy consumption and utilization rate based on the selected conditions to generate and display the reports.

[0037] The system management module includes a user management module and a permission management module. The user management module records and modifies user information. Designed for ordinary users, it records basic user data, including nicknames, avatars, and other basic account-related information. The permission management module, designed for administrators, sets account permissions and configures account roles. Its function is to restrict platform account permissions and configure account roles.

[0038] Real-time monitoring modules include map positioning and real-time video, providing users with real-time and visualized equipment monitoring status, allowing users to intuitively, remotely, and quickly grasp the equipment's operating status.

[0039] Expert systems are a core function of the business application layer. Composed of modules such as fault diagnosis and lifespan prediction, expert systems support intelligent operation and maintenance. The expert system module extracts features from equipment operating data, which serve as input for model inference. The model then assesses the health status of each piece of equipment based on the operating data, i.e., fault diagnosis and lifespan prediction. Based on the assessment results, the platform can adjust production equipment, achieving automatic optimization and intelligent scheduling of the production process.

[0040] Specifically, such as Figure 2 As shown, the workflow of the expert system module is as follows: It acquires heterogeneous data after processing by the data communication layer, including vibration signals, temperature signals, current signals, etc., to overcome the limitations of single signals, so as to comprehensively reflect the health status of the equipment and improve the accuracy and reliability of diagnosis.

[0041] Heterogeneous data undergoes preprocessing such as cleaning and denoising, followed by feature extraction. These features include time-domain features, frequency-domain features, and time-frequency-domain features. These three types of features describe signals from different dimensions, complementing each other and avoiding the limitations of single features. Time-domain features primarily reflect the overall state of the equipment, including statistical characteristics, axis trajectory, and features related to chaos and entropy, which helps monitor equipment health and predict fault trends. Frequency-domain features are extracted using methods such as Fourier transform, holographic spectrum technology, and higher-order spectral analysis to determine the specific type, cause, and location of faults. Time-frequency-domain features are extracted using methods such as short-time Fourier transform, wavelet analysis, and empirical mode decomposition, suitable for analyzing signals during equipment startup, shutdown, or unstable operation.

[0042] The extracted features are input into a trained deep learning model for equipment fault diagnosis or lifespan prediction. The core of the expert system is to construct an intelligent model based on deep neural networks and integrating traditional machine learning methods. For example, clustering algorithms, convolutional neural networks, and empirical mode decomposition are used for fault diagnosis; hidden Markov models, Gaussian process regression, correlation vector machines, and recurrent neural networks are used for lifespan prediction. This deep learning model aims to fully leverage the powerful capabilities of deep learning in processing high-dimensional, nonlinear time-series data, automatically extracting deep features from multi-source signals (such as vibration, temperature, and current) of the equipment. Through advanced feature fusion technology, features from different sources are organically integrated and collaboratively analyzed within the model, thereby achieving accurate identification of complex faults and reliable prediction of the remaining lifespan of the equipment. This strategy ensures that the expert system can both detect subtle fault symptoms and grasp macroscopic performance degradation trends.

[0043] Furthermore, such as Figure 4 As shown, the training process for the above deep learning model is as follows: Step 1: Acquire raw signals from multiple sources in industrial equipment. By deploying multi-source sensors at key locations on the industrial equipment, raw signal data during equipment operation is acquired in real time. The aim is to overcome the limitations of single signals, comprehensively capture multi-dimensional information about the equipment's operating status, and provide realistic and comprehensive raw input data for subsequent model training.

[0044] Step 2: Preprocessing the raw signals. Systematic preprocessing is performed on the collected multi-source raw signals, specifically including data cleaning to remove invalid data and ensure data reliability; data normalization to eliminate dimensional differences between different signals and ensure balanced feature weights during model training; and data augmentation to expand the dataset and improve the model's adaptability to data variations.

[0045] Step 3: Divide the dataset into training, validation, and test sets. Divide the preprocessed dataset into three subsets in a ratio of 7:2:1 (training:validation:test). The training set is used for parameter learning of the neural network model, adjusting the model's weights and biases using backpropagation. The validation set is used for performance monitoring during training, tracking the model's generalization ability on unseen data and preventing overlearning. The test set is used for final evaluation of the model's generalization performance, independent of the training and validation sets, ensuring the objectivity of the evaluation results.

[0046] Step 4: Construct a dedicated neural network model for industrial time-series data. To address the temporal nature (e.g., signal trends over time) and local features (e.g., impact pulses caused by faults) of industrial equipment operation data, a multi-source parallel convolutional-temporal-dependency capture-attention mechanism weighted fusion neural network model was constructed, as follows: Figure 5 As shown, the neural network model includes: Multi-source independent convolutional layers design convolutional kernels of different sizes for each signal branch. Small convolutional kernels capture high-frequency impacts caused by faults, while large convolutional kernels capture low-frequency trends of lifetime degradation. Independent convolution ensures that the unique physical characteristics of each signal are not confused.

[0047] Pooling layers reduce the dimensionality of the output of multi-source independent convolutional layers to condense key features and reduce computation. This operation can accurately capture key transients in the signal. For fault diagnosis, it captures the "moment when the fault features are most obvious"; for lifetime prediction, it captures the "inflection point of accelerated performance degradation".

[0048] Bidirectional LSTM layers can capture long-term temporal dependencies, understand how states change over time, and achieve "cause tracing" for fault diagnosis and "future evolution" for lifetime prediction.

[0049] The attention mechanism layer weights temporal features, filters redundant information, and identifies features at "critical moments," thereby improving the effectiveness of features and the interpretability of the model.

[0050] Step 5: Set training parameters and start model training. Select the Adam optimizer and set appropriate parameters such as the number of iterations and batch size; determine the loss function according to the task type, using the cross-entropy loss function for fault diagnosis tasks and the mean squared error loss function for life prediction tasks; input the training set data into the model in batches, calculate the prediction results through forward propagation, and then use backpropagation to bias the network weights to enable the model to learn the patterns in the data.

[0051] Step Six: Track Validation Set Metrics to Avoid Overfitting. After each iteration, evaluate model performance using validation set data: for fault diagnosis tasks, track validation set accuracy; for lifetime prediction tasks, track validation set root mean square error (RMSE). If the validation set metric does not improve for three consecutive iterations, stop training early to prevent the model from overlearning noise from the training set.

[0052] Step 7: Apply optimization techniques to improve model performance. For cases where validation set performance is substandard, methods such as regularization, Dropout, and hyperparameter tuning can be used to optimize the model. Regularization adds a regularization term to the loss function to limit the absolute values ​​of model parameters and reduce model complexity; Dropout adds a Dropout layer between the CNN and LSTM layers, randomly dropping some neurons to prevent the model from over-relying on specific features; and grid search is used to adjust hyperparameters such as convolutional kernel size and the number of hidden units to optimize the network structure and improve training stability.

[0053] Step 8: Evaluate and iteratively adjust model performance. After training, independently evaluate the model using test set data: for fault diagnosis tasks, calculate test set accuracy, precision, and recall; for lifetime prediction tasks, calculate test set root mean square error (MSE) and mean absolute error (MAE). If the test set metrics do not meet expectations: if the problem is attributed to the model architecture, return to Step 4 to reconstruct the network; if the problem is attributed to the training process, return to Step 5 to adjust parameters; if the problem is attributed to the data foundation, return to Step 2 to optimize the data and features. If the metrics meet expectations, save the model's optimal parameters and complete model training.

[0054] Step Nine: Solidify and Output the Model. Solidify the model parameters that have achieved the expected goals, save the trained model, and convert the model into a lightweight format suitable for industrial environment deployment to reduce model size, improve inference speed, and use it for actual fault diagnosis or life prediction tasks, providing support for intelligent operation and maintenance of industrial equipment.

[0055] It should be noted that the present invention is not limited to the above-described embodiments. The above embodiments are merely examples, and any embodiments that have the same structure and perform the same effects as the technical concept within the scope of the present invention are included within the scope of the present invention. Furthermore, various modifications that can be conceived by those skilled in the art to the embodiments, and other ways of constructing by combining some of the constituent elements of the embodiments, without departing from the spirit of the present invention, are also included within the scope of the present invention.

Claims

1. A smart operation and maintenance platform based on deep learning and industrial Internet of Things, characterized in that, include: The equipment layer is used to collect operational data of production equipment through a variety of sensors and actuators integrated in the industrial field; The acquisition and transmission layer is used for preliminary processing of the acquired data; The data communication layer is used to integrate and standardize the data after preliminary processing. The business application layer provides operation and maintenance management functions based on the data processed by the data communication layer. It includes an expert system module, which uses a deep learning model to assess the health status of the equipment based on the equipment operation data.

2. The intelligent operation and maintenance platform based on deep learning and industrial IoT as described in claim 1, characterized in that, The workflow of the expert system module is as follows: Acquire heterogeneous data after it has been processed by the data communication layer; Heterogeneous data is preprocessed and features are extracted, including time-domain features, frequency-domain features, and time-frequency-domain features. The extracted features are input into a trained deep learning model for equipment fault diagnosis or lifespan prediction.

3. The intelligent operation and maintenance platform based on deep learning and industrial IoT as described in claim 2, characterized in that, The deep learning model is a neural network model that combines multi-source parallel convolution, temporal dependency capture, and attention mechanisms in a weighted manner.

4. The intelligent operation and maintenance platform based on deep learning and industrial IoT as described in claim 3, characterized in that, The training process of the deep learning model includes: The preprocessed heterogeneous data is divided into training set, validation set and test set, and the neural network model is constructed accordingly. The model was trained using the Adam optimizer, and an early stopping strategy was employed to prevent overfitting. The model is tuned using the validation set and its performance is evaluated using the test set.

5. The intelligent operation and maintenance platform based on deep learning and industrial IoT as described in claim 1, characterized in that, The deep learning model includes: Multi-source independent convolutional layers are used to process signals from different sources using convolutional kernels of different sizes, in order to extract local features of different signals; Pooling layers are used to reduce the dimensionality of the outputs of multi-source independent convolutional layers to condense key features; A bidirectional LSTM layer is used to capture long-term temporal dependencies of device states and output temporal features. An attention mechanism layer is used to weight the temporal features to focus on features at key moments.

6. The intelligent operation and maintenance platform based on deep learning and industrial IoT as described in claim 5, characterized in that, The process of using convolutional kernels of different sizes to process signals from different sources specifically involves: Small convolutional kernels are used to capture high-frequency impact features caused by faults, while large convolutional kernels are used to capture low-frequency trend features of lifetime degradation.

7. The intelligent operation and maintenance platform based on deep learning and industrial IoT as described in claim 1, characterized in that, The business application layer also includes: The equipment management module is used for equipment information entry, historical equipment data query, and production process management. The alarm management module is used to receive abnormal information output by the expert system module and issue fault alarms, while also providing query functions for fault information and platform historical events. The report management module is used to automatically generate energy consumption reports and equipment utilization reports based on the conditions selected by the user. The system management module is used to manage the basic information of platform users and configure access and operation permissions for different accounts.

8. The intelligent operation and maintenance platform based on deep learning and industrial IoT as described in claim 1, characterized in that, The data communication layer includes: The core bus is used for aggregating, standardizing, and distributing heterogeneous data in real time. Distributed data communication systems are deployed in factory workshops or at the edge of production lines to collect and preprocess data locally.

Images