A cloud edge end cooperative hot rolling process hydraulic system oil leakage monitoring system

Abstract

The application discloses a kind of cloud edge end collaborative hot continuous rolling process hydraulic system oil leakage monitoring system, belong to industrial fault monitoring technical field, the system uses cloud edge end architecture;Wherein, end side is used to collect the working parameter of hot continuous rolling process finishing mill hydraulic screwdown system;Edge side is used to the working parameter of end side collection is preprocessed, and the working parameter after preprocessing is input into the detection model trained by cloud, utilizes the detection result output by the detection model, to detect whether the hot continuous rolling process finishing mill hydraulic screwdown system appears oil leakage fault;Wherein, detection model fuses convolutional neural network, long short-term memory network and multi-head attention mechanism;Cloud is used for the training and updating of detection model.The application uses deep learning fusion model to carry out feature extraction to multi-source time series signal, intelligently identifies to leakage type, and can realize the rapid early warning and visual display of hydraulic system oil leakage.

Description

A cloud-edge-end collaborative hydraulic system for monitoring oil leakage in hot continuous rolling processes. Technical Field

[0001] This invention relates to the field of industrial fault monitoring technology, and in particular to a cloud-edge-end collaborative hydraulic system for monitoring oil leakage in hot continuous rolling processes. Background Technology

[0002] Hydraulic systems primarily utilize the pressure energy of fluid within a hydraulic circuit to transmit force and motion. They mainly consist of five parts: power source, actuators, control components, auxiliary components, and working medium. The working medium within a hydraulic system can, due to its fluid properties, transmit energy almost without loss between two points over long distances, generating stable, high thrust or torque. Hydraulic systems also feature rapid start-up, braking, and speed change capabilities, making them widely used in large and specialized equipment.

[0003] Hot strip rolling is a highly efficient metal processing technology that shapes metal billets into desired specifications through rolling mills. The hydraulic system plays a crucial role in this process. Throughout the hot strip rolling process, the reduction hydraulic system is responsible for adjusting the roll gap, thereby controlling the strip thickness. In hot strip rolling, the reduction adjustment primarily relies on the AGC (Automatic Gauge Control) system, which controls the roll gap to achieve strip thickness control. Bending rolls and shifting rolls are mainly used to control the strip's crown and straightness, and these are also controlled by the hydraulic system.

[0004] The hydraulic system in hot continuous rolling operates under high temperature and pressure for extended periods. This makes the interfaces of the hydraulic circuits prone to loosening, and the seals of the hydraulic cylinders susceptible to aging and deformation, leading to internal or external leaks. Hydraulic system leaks are relatively slow, making them difficult to detect in their early stages. However, the impact of leaks on the system is significant. Leaks can cause: reduced hydraulic circuit pressure, preventing the hydraulic cylinders from outputting target force or torque in a timely manner, thus affecting the accuracy of functions such as pressing and bending rolls, ultimately resulting in uneven thickness, warping, and surface defects in the rolled slab; oil leaks can also lead to insufficient hydraulic oil in the hydraulic pump and other critical components; and leaked hydraulic oil can contaminate the production environment, negatively impacting personnel health and the working environment. Therefore, monitoring hydraulic system leaks has become a crucial and unavoidable issue in hot continuous rolling production.

[0005] Specifically, the main oil leakage faults in the hydraulic system are as follows:

[0006] Regarding hydraulic cylinder oil leakage, aging or damage to the sealing piston between the working chamber and the back pressure chamber can cause internal leakage of oil from the piston, i.e., hydraulic oil leaks from the high-pressure chamber to the low-pressure chamber. This not only causes oil leakage but also leads to unstable cylinder movement speed and insufficient thrust. A bent or deformed piston rod will cause abnormal friction and compression between the piston rod and the rod-side sealing element during movement, resulting in excessive wear of the sealing element and thus oil leakage. Problems during hydraulic cylinder installation, such as incorrect installation position, excessive or insufficient installation torque, or sealing elements that are installed too loosely or too tightly, can lead to uneven stress on internal components, causing the sealing elements to malfunction or even damaging the connection points, ultimately resulting in oil leakage.

[0007] During hot continuous rolling, hydraulic pump leakage is a significant concern. This type of failure can be caused by various factors, including a low oil level in the tank, blockage in the suction pipe, gas leaks, or a clogged filter at the hydraulic pump's suction port. When the oil level is low, the pump's suction process becomes difficult, resulting in insufficient oil intake, poor internal lubrication, accelerated wear of internal parts, and ultimately, leakage. Internal leaks or blockages in the hydraulic pump will reduce its output oil volume, causing the relief valve to be set too high and increasing system pressure. Poor cooler performance leads to increased hydraulic oil temperature and decreased viscosity, also increasing leakage and further raising system pressure. Secondly, when impurities accumulate in the oil and clog the delivery pipe, the internal pressure of the hydraulic pump becomes unbalanced, damaging sealing elements under abnormal pressure and ultimately causing leakage. Furthermore, gas leaks near the hydraulic pump's suction port can interfere with normal pump operation. Unlike oil, gas is not compressible. When gas enters the pump, it alters the pump's volumetric efficiency, leading to increased wear on internal components. Furthermore, gas leakage disrupts the oil pressure balance around sealing elements, damaging seals and causing oil leaks.

[0008] Servo valve malfunction is another potential problem in the hydraulic system of hot strip milling. Damage or wear to the sealing surface can cause oil leakage from the gap between the valve core and the valve body, reducing the control accuracy of the hydraulic valve and affecting the operation of the entire hydraulic system. Secondly, wear, aging, corrosion, or inherent quality problems in the pipeline caused by impurities, as well as loosening or improper installation of joints due to mechanical vibration, can also lead to oil leakage. In addition, blockage and contamination are also common causes of servo valve oil leakage. Impurities and particles in the oil can easily clog the orifices and gaps of the servo valve. After long-term use, the high precision of the original fit between the valve core and the valve sleeve will increase due to wear, causing the valve core to malfunction, jam, or experience a decrease in flow and pressure control accuracy, increased internal leakage, and affecting the stability and response speed of the system.

[0009] Finally, back pressure side leakage is also a significant aspect of hydraulic system oil leakage. Aging, wear, or damage to the seals of the back pressure valve can lead to oil leakage from the back pressure side. Improper adjustment of the back pressure valve, such as setting the back pressure too low, can also cause leakage due to pressure differential. Furthermore, loose connections or seal failure in the back pressure side piping can also cause oil leakage. Back pressure side leakage leads to unstable system pressure, affecting the normal operation of actuators and reducing the efficiency and reliability of the hydraulic system.

[0010] Today, downstream enterprises have higher requirements for the quality of high-end strip steel products. Timely detection and resolution of hydraulic system oil leaks can help improve production efficiency, enhance product quality, and achieve cost reduction and efficiency improvement. Traditional oil leak monitoring mainly relies on experienced operators conducting regular inspections, subjectively judging the presence of leaks through manual observation and their own experience. However, this traditional manual inspection method is slow and cannot detect early-stage leaks in time, resulting in low efficiency. There are also methods based on signal processing or shallow machine learning, but these mostly process or build models based on simulated data of hydraulic systems, which differ from the actual production process, and the models built are relatively simple. With the development of computer technology and artificial intelligence technology, data analysis methods have been applied in various industries. This method uses computer technology to analyze actual data during equipment operation, while building various models to achieve multiple objectives, including fault diagnosis. As can be seen from the survey of hydraulic system fault diagnosis methods, although data-driven fault diagnosis methods have shown great potential in the field of hydraulic system fault diagnosis, existing research on oil leakage monitoring in hydraulic systems of hot rolling processes is still mainly focused on signal processing, statistical analysis or shallow machine learning. These methods have certain limitations when dealing with complex nonlinear relationships and high-dimensional data. Summary of the Invention

[0011] This invention provides a cloud-edge-end collaborative hydraulic system for monitoring oil leakage in hot continuous rolling processes, in order to solve the technical problem that existing technologies have certain limitations when dealing with complex nonlinear relationships and high-dimensional data.

[0012] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0013] On one hand, this invention provides a cloud-edge-end collaborative hydraulic system for monitoring oil leakage in a hot continuous rolling process. The system adopts a cloud-edge-end architecture, including a cloud layer, an edge layer, and an end layer.

[0014] The end side is used to collect the working parameters of the hydraulic reduction system of the finishing mill in the hot continuous rolling process;

[0015] The side is used to preprocess the working parameters collected from the end side, and input the preprocessed working parameters into the detection model trained in the cloud. The detection model is used to output the detection results to detect whether there is an oil leakage fault in the hydraulic pressing system of the finishing mill in the hot continuous rolling process. The detection model integrates convolutional neural network, long short-term memory network and multi-head attention mechanism of Transformer.

[0016] The cloud is used for training and updating the detection model.

[0017] Furthermore, the operating parameters include the pressure, position, and flow rate of the hydraulic cylinder back pressure chamber and working chamber of each stand in the hot continuous rolling mill, as well as the servo valve current of each stand's hydraulic system.

[0018] Furthermore, the preprocessing includes:

[0019] Standardize the working parameters;

[0020] Windowing is applied to the standardized working parameters.

[0021] Furthermore, the output of the detection model is the type of oil leakage in the hydraulic system during the hot rolling process; among which, the types of oil leakage in the hydraulic system include: no leakage, leakage inside the hydraulic cylinder, leakage on the back pressure side, and leakage inside the servo valve.

[0022] Furthermore, the detection model includes a first feature extraction module, a second feature extraction module, and an attention module; wherein, the second feature extraction module includes a CNN network and an LSTM network; and the attention module adopts the Transformer multi-head attention mechanism.

[0023] The detection model's processing of input data includes:

[0024] First, the input data is divided into multiple samples by a sliding window. Then, the first feature extraction module performs one-dimensional convolution, batch normalization and max pooling operations on each sample to extract the first data feature. The first data feature is then input into the second feature extraction module and the attention module respectively.

[0025] The second feature extraction module first uses a CNN network to perform convolution, batch normalization, and max pooling operations on the first data features to extract the second data features. Then, the second data features are input into an LSTM network to extract the third data features.

[0026] The attention module uses multiple heads to focus on different aspects of the first data feature and capture the global correlation between different features in the data to extract the fourth data feature;

[0027] The third and fourth data features are fused to obtain the fused features; the fused features are mapped to the classification space through a fully connected layer to distinguish between the normal state and different types of oil leaks.

[0028] Furthermore, the loss function of the detection model adopts a weighted combination of cross-entropy and L2 regularization.

[0029] Furthermore, the detection model employs an early stopping strategy and adaptive learning rate scheduling during training.

[0030] Furthermore, the attention module uses query-key interaction calculations with four independent attention heads to uncover implicit correlations between different time steps, and the attention weight matrix of the multi-head attention mechanism is normalized by the LayerNormalization layer; the dropout rate is set to 0.25 to prevent overfitting.

[0031] In another aspect, the present invention also provides an electronic device comprising a processor and a memory; wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the above-described method.

[0032] In another aspect, the present invention also provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the above method.

[0033] The beneficial effects of the technical solution provided by this invention include at least the following:

[0034] This invention deploys models on the edge, trains and manages models in the cloud, and completes data collection and visualization on the device side. This enables the system to quickly and accurately identify and classify hydraulic system oil leakage conditions, even in scenarios with complex operating conditions, multiple noise superpositions, and multivariate coupling. Through a cloud-edge-device collaborative mechanism, this invention achieves hierarchical management of computing tasks and closed-loop management of information flow, ensuring both the real-time performance and stability of hydraulic system oil leakage detection, while also considering network resource utilization efficiency and system security. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 is a schematic diagram of the CNN-LSTM-MultiHeadAttention fusion model structure provided in an embodiment of the present invention;

[0037] Figure 2 shows the training accuracy and validation accuracy of the CNN-LSTM-MultiHeadAttention model provided in the embodiment of the present invention, as well as the training loss and validation loss; wherein, (a) is the training accuracy and validation accuracy of CNN-LSTM-MultiHeadAttention; and (b) is the training loss and validation loss of CNN-LSTM-MultiHeadAttention.

[0038] Figure 3 is the confusion matrix of the CNN-LSTM-MultiHeadAttention test results provided in the embodiment of the present invention;

[0039] Figure 4 is a cloud-edge-device collaborative system architecture diagram provided in an embodiment of the present invention;

[0040] Figure 5 is a microservice architecture diagram based on Flask and Django provided in an embodiment of the present invention;

[0041] Figure 6 is a schematic diagram of the system front-end monitoring and alarm interface provided in an embodiment of the present invention; wherein, (a) is the system homepage; (b) is the real-time diagnosis and abnormal situation record; and (c) is the query of abnormal status within the corresponding time period;

[0042] Figure 7 is a system block diagram of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0044] First, it should be noted that in the embodiments of the present invention, the words "exemplarily," "for example," etc., are used to indicate that they are examples, illustrations, or descriptions. Any embodiment or design scheme described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the term "exemplarily" is intended to present the concept in a specific manner. Furthermore, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either one or the other.

[0045] First Embodiment

[0046] To address the problems of large detection latency, insufficient accuracy and robustness, and high engineering deployment costs in existing technologies, this embodiment provides a cloud-edge-device collaborative oil leakage monitoring system for hot continuous rolling mill hydraulic systems. It proposes a fusion method that utilizes actual production data from the hot continuous rolling mill hydraulic system, combined with a multi-head attention mechanism using convolutional neural networks, long short-term memory networks, and Transformer, to monitor oil leakage in the mill's roll-down hydraulic system. An oil leakage fault monitoring model is constructed based on actual hot continuous rolling process data, and the model is tested using real data. Furthermore, multi-sensor data fusion enhances the model's robustness, making it perform well in noise resistance and adaptability to complex operating conditions.

[0047] The system adopts a cloud-edge-device architecture, including cloud, edge, and device: where,

[0048] The end side is used to collect the working parameters of the hydraulic pressing system of the finishing mill in the hot continuous rolling process; among them, the working parameters include the pressure, position and flow of the hydraulic cylinder back pressure chamber and working chamber of each stand of the finishing mill in the hot continuous rolling process, as well as multi-source data such as the servo valve current of each stand's hydraulic system.

[0049] The edge side is used to preprocess the working parameters collected from the end side, and the preprocessed data is input into the detection model trained in the cloud. The detection model outputs the detection results to detect whether there is an oil leakage fault in the hydraulic pressing system of the finishing mill in the hot continuous rolling process. The preprocessing includes standardizing the data and windowing the standardized data. The output of the detection model is the type of oil leakage in the hydraulic system of the hot continuous rolling process. The types of oil leakage in the hydraulic system include no leakage, leakage inside the hydraulic cylinder, leakage on the back pressure side, and leakage inside the servo valve. The detection model integrates the multi-head attention mechanism of convolutional neural network, long short-term memory network and Transformer, denoted as CNN-LSTM-MultiHeadAttention, and its structure is shown in Figure 1.

[0050] The cloud is used for training and updating the detection model.

[0051] Specifically, the system construction process of this embodiment is as follows:

[0052] I. Data Acquisition and Processing

[0053] First, based on the mechanism of the hydraulic reduction system of the finishing mill in the hot strip mill process, relevant variables of the hydraulic system were screened, and leakage-related data were identified. A variable template was created based on the screened variables. Using the created template, relevant production data from seven finishing mills (F1 to F7) in the hot strip mill process of a steel plant's 2150 production line were examined. Three types of hydraulic system oil leakage faults were found in the existing data: internal leakage in the hydraulic cylinder, back pressure side leakage, and internal leakage in the servo valve. These leaks mainly occurred in the four stands (F1 to F4).

[0054] 1. Internal leakage in hydraulic cylinders: In the hot continuous rolling process, the hydraulic cylinders of the roll pressing hydraulic system mainly control the rise and fall of the rolls by the pressure difference between the back pressure chamber and the working chamber. Internal leakage in the hydraulic cylinders is mainly related to the seals between the two chambers. After working under high pressure and high intensity, the seals between the two chambers are prone to aging and wear. This leakage will cause hydraulic oil in the high-pressure chamber to leak into the low-pressure chamber during the rolling process, resulting in a decrease in the pressure difference between the two sides, which in turn affects the rolling force.

[0055] 2. Back pressure side leakage: The hydraulic system is equipped with a back pressure valve or back pressure circuit in the back pressure chamber to maintain a certain pressure to prevent cavitation or improve stability. If a leak occurs in the back pressure circuit (such as a pipe rupture, loose joint, or failure of the back pressure valve), the back pressure side pressure will drop because it cannot be maintained.

[0056] 3. Internal leakage of servo valve: When the clearance between the servo valve core and the valve sleeve increases, even if the valve core is in the neutral position (not activated), high-pressure oil will still leak through the clearance to the return port, causing system pressure fluctuations. At this time, the controller needs to frequently adjust the output to maintain the target pressure, which causes the servo valve current feedback display to fluctuate frequently.

[0057] Then, actual production history data from four stands (F1 to F4) of the finishing mill in the 2150 production line of the steel plant were extracted, totaling 86 slab data points. Of these, 53 slab data points were divided into training and validation sets, and the remaining 33 were divided into a test set. The final results of the division are detailed in Table 1. Each extracted slab data point has approximately 5000-15000 time steps, with the number of time steps varying for different slabs.

[0058] Table 1 Dataset partitioning results

[0059]

[0060] Before training, validation, or testing, all data from the corresponding dataset is loaded into a list. Then, the `vstack` array concatenation method of NumPy arrays is used to vertically stack the data in the list into a large array. Finally, the data in the array is standardized using `StandardScale`, ensuring that the mean of each of the 23 features in the preprocessed data is 0 and the standard deviation is 1. The transformation function is:

[0061]

[0062] in, The mean of all sample data. Let be the standard deviation of all sample data. After standardization, time windows need to be created for the standardized data, with each time window representing one sample. The size of the time window is Window_size, and the step size is Step_size.

[0063] II. Construction and Training of the Fusion Model

[0064] This embodiment employs a deep fusion model of Convolutional Neural Network (CNN), Long Short-Term Memory Network (LSTM), and MultiHeadAttention to capture local spatial features and long-term dependencies in sensor signals. Simultaneously, it uses an attention mechanism to adaptively weight and focus on key leaked features, significantly improving generalization ability and anti-interference capability under various conditions. The model input is a multi-channel time-series tensor. First, local patterns are extracted through several convolutional layers. Then, the convolutional features are fed into the LSTM layer in chronological order for state updates. The core of LSTM lies in achieving joint modeling of short-term and long-term dependencies through forget gates, input gates, and output gates. Its calculation process is as follows:

[0065]

[0066]

[0067]

[0068]

[0069]

[0070]

[0071]

[0072] in, , and These represent the forget gate, input gate, and output gate, respectively. For the state of the memory unit, This is the current hidden state. The structure dynamically adjusts the information retention and update ratio over time using a gating mechanism, enabling long-term dependency modeling of hydraulic signals.

[0073] Furthermore, in fault diagnosis tasks involving long-term time-series data, even though LSTM adds gating mechanisms and cellular structures to the RNN (Recurrent Neural Network) Encoder-Decoder model, allowing LSTM to selectively forget and remember, the model may still fail to correlate fault states over long timelines. Therefore, this embodiment introduces the MultiHeadAttention mechanism from the Transformer algorithm on top of the CNN-LSTM algorithm. This leverages the model's ability to focus on different aspects of information from multiple heads, capturing important information from long-term time series.

[0074] Therefore, after obtaining the hidden state sequence, a multi-head attention mechanism is introduced to enhance global feature selection and weighted fusion. The attention layer first calculates the query vector. Key vector AND value vector The correlation between them, expressed by the scaled dot product attention function, is as follows:

[0075]

[0076] The multi-head mechanism learns features of different subspaces in parallel through multiple independent attention heads, specifically:

[0077]

[0078]

[0079]

[0080] This mechanism enables weighted fusion of features from multiple time steps and channels globally, allowing the model to automatically focus on key signal segments related to leaks. The final output features are mapped to the classification space via a fully connected layer to distinguish between normal states and different types of oil leaks. During training, weighted cross-entropy is used to balance the number of samples for different leak categories, while L2 regularization and early stopping strategies are combined to avoid overfitting.

[0081] The model's processing of input data includes:

[0082] The model training code receives 23-dimensional sensor signals from 53 slab data points. Since the data within each of the 53 slabs represents the same health state, each slab's data corresponds to a single label. The label file only needs to list the slab file name and the corresponding health state. First, the slab and label files are loaded. The column names in the label file are mapped to the slab file names. Then, the label names corresponding to the rows in the label file associated with the slab file names are mapped to each time step in the slab file. The loaded data is then normalized using the StandardScale method, setting the mean of each channel signal to zero and the variance to one.

[0083] The standardized time-series signal is segmented using an overlapping sliding window method, dividing the input data into multiple samples. Each sliding window has 10 time steps and a sliding step size of 5 time steps, ensuring 50% data overlap between adjacent windows. This maintains the continuity of local time segments while enhancing model robustness in small-sample scenarios through data redundancy. Each sliding window represents one sample and can only have one label; therefore, the label of each window is determined by the label of the last time step within the window. After this processing, the original one-dimensional sequence is reconstructed into a three-dimensional tensor (number of samples, 10, 23) to meet the requirements of deep learning models for joint extraction of spatiotemporal features.

[0084] The reconstructed 3D tensor is input into the function of the CNN-LSTM-MultiHeadAttention model. First, the input data undergoes one-dimensional convolution (conv1d), batch normalization (BatchNormalization), and max pooling (MaxPooling1D) to extract data features and enhance the robustness of the extracted features. Convolution1d not only automatically extracts local features from the input data but also, in conjunction with convolutional and pooling layers, reduces the dimensionality of the data while preserving important feature information, thus reducing the feature dimension and computational cost. The features extracted by the first CNN layer are input into the CNN-LSTM layer, where the CNN layer repeats the feature extraction process. After BatchNormalization and MaxPooling1D, the output is sent to the LSTM layer. The LSTM then captures the temporal relationship between the features after the second dimensionality reduction. Specifically, in this embodiment, a single 128-neuron LSTM layer captures the temporal information of the features extracted by the two convolutional operations, while setting return_sequences=False to represent the overall state of the focusing window.

[0085] The multi-head attention mechanism also receives key-value pairs extracted from the output of the first-layer CNN. This layer utilizes information from different aspects of the features focused on by multiple heads and captures global correlations between different features in the data. Specifically, in this embodiment, this layer mines implicit correlations between different time steps through query-key interaction calculations by four independent attention heads, enabling the model to decouple the discriminative features of multiple fault modes. Furthermore, the attention weight matrix of the multi-head attention mechanism is normalized by the LayerNormalization layer, alleviating the gradient vanishing problem, and the Dropout rate of 0.25 also prevents overfitting to some extent.

[0086] The 128-dimensional global temporal features output by the LSTM layer are tensor-concatenated with the local correlation features flattened by the attention module, which is fused in the feature extraction layer to form a high-dimensional joint representation space, enabling the model to generate more comprehensive feature representations.

[0087] The classifier module employs a strategy combining fully connected layers and regularization techniques: the 256-neuron hidden layer is equipped with ReLU activation and L2 regularization constraints (λ=0.001), enhancing nonlinear mapping capabilities while controlling model complexity; the output layer's Softmax function projects the feature space onto four probability distributions, and the cross-entropy loss function guides the model optimization direction. The training process also utilizes a dynamic learning strategy: the Adam optimizer's initial learning rate is set to 0.0001, with gradient clipping (threshold 1.0) to prevent gradient explosion; the ReduceLROnPlateau callback function monitors and verifies loss changes; after 5 rounds of loss plateau, the learning rate decays by 50%; and the EarlyStopping mechanism terminates training after 10 rounds without improvement, ensuring the model converges to the region of optimal generalization performance.

[0088] The overall model framework is shown in Figure 1. In this model, the convolutional module is used to extract local spatial features from the multi-channel time series of hydraulic signals. Multi-scale feature representations are obtained through multi-layer convolution and pooling operations to capture leakage features under different frequencies and amplitude variations. The LSTM module uses a gating structure to dynamically update the input signal, simultaneously representing short-term fluctuations and long-term trends, maintaining the model's continuity and stability in the time dimension. The multi-head attention mechanism is used to adaptively allocate feature weights globally, enabling the model to automatically focus on key time slices and channels related to oil leakage, thereby enhancing the ability to identify complex coupled features. Differential attention to different time slices and feature channels is achieved through parallel modeling of multiple query-key-value subspaces. In the loss function design, a weighted combination of cross-entropy and L2 regularization is used, supplemented by an early stopping strategy and adaptive learning rate scheduling to suppress overfitting and accelerate convergence.

[0089] III. System Setup

[0090] Furthermore, to ensure the trained model can be applied in practice, this embodiment comprehensively utilizes multiple technologies, including Python, MySQL database, Django framework, Flask framework, FastAPI framework, Docker container technology, JavaScript technology, and cloud-edge-device architecture, to develop a hot strip mill hydraulic system leakage monitoring system based on a B / S architecture. This system possesses three functions: real-time mill status monitoring, model training and testing, and data management. Considering the non-stationarity and distributed drift in the hot strip mill environment, this system supports a continuous learning and hot model update mechanism through edge-cloud collaboration, enabling rapid adaptation of incremental data without affecting production. Simultaneously, to reduce diagnostic latency, improve data transmission security, and enhance overall system performance, the training and testing code is separated from the Django framework using the Flask framework. Subsequently, the Flask framework is deployed on a cloud server, leveraging the powerful computing capabilities of the cloud to support complex training tasks; the Django framework and FastAPI service are deployed on an edge server, bringing the fault diagnosis core closer to the edge, reducing diagnostic latency, minimizing network overhead, and ensuring the overall performance and reliability of the system. Finally, Pspace is used to call back historical data to simulate real-time data acquisition during production, thereby testing the mill's real-time status monitoring function. In terms of system architecture, this embodiment establishes a three-level collaboration between cloud, edge, and terminal: the cloud provides model training, evaluation, and parameter management services through Flask, aggregating and uniformly optimizing data from multiple production lines and shifts; the edge side uses FastAPI to carry inference services and lightweight caching, performing leak identification and type judgment on real-time data with millisecond-level response; terminal devices collect multi-source signals such as pressure, position, flow rate, and servo valve current from stands F1 to F7, and use pSpace for historical data playback and simulation. All three are modularly deployed and elastically expanded through Docker containerization, and model parameters, logs, and alarm information are uniformly stored through MySQL, thus achieving a highly reliable, maintainable, and easily migrated engineering implementation.

[0091] Based on the above, the cloud-edge-device collaborative mechanism constructed in this embodiment aims to achieve hierarchical processing and efficient collaboration of hydraulic system oil leakage monitoring tasks, thereby taking into account the system's real-time performance, reliability, and maintainability. The entire system consists of cloud-side, edge-side, and device-side components, which form a complete closed loop of acquisition, transmission, analysis, diagnosis, and feedback through the collaborative action of data flow and control flow.

[0092] The cloud side primarily handles global and complex computational tasks, serving as the core data and model management center for the entire system. It undertakes model training, testing, and updating, utilizing historical hydraulic signal data for iterative model optimization. After training, the system automatically synchronizes the updated model weight file to the edge nodes, ensuring the on-site diagnostic model remains up-to-date. The cloud also stores multiple batches of model training results and performance metrics for subsequent comparison and verification. This layer also stores system operation logs, alarm records, and user configuration parameters, providing model access and status query services to the edge side via a backend interface. Centralized management in the cloud enables unified scheduling and remote updates of multiple production lines and hydraulic stations, reducing on-site maintenance workload and improving overall system controllability.

[0093] Located on a local server or industrial computer near the rolling mill, the edge node is the core of real-time diagnostics and user interaction. It primarily receives hydraulic signal data transmitted from the end-side equipment, caches, preprocesses, and extracts features from the data, and performs model inference and status judgment locally. The edge node deploys the Django framework to handle backend logic and user page display, while utilizing FastAPI to provide high-speed data interface services for multi-threaded data processing and rapid response. This layer can perform parallel calculations and visualization of real-time collected hydraulic pressure, oil temperature, displacement, and current signals. When abnormal features are detected, the system will immediately alert the user interface via an alarm window or color change, while recording the alarm time, location, and corresponding signal trend. The edge also features data management and local storage capabilities, saving historical data segments according to user-defined time windows, supporting comparative analysis and visual playback of historical signals. Furthermore, the edge acts as a data relay, uploading filtered and compressed key information to the cloud to provide input for subsequent model optimization and statistical analysis. By completing the main data processing and inference tasks at the edge, the latency and bandwidth consumption caused by directly uploading large amounts of raw data are avoided, enabling the system to maintain high response speed and stable performance even in complex network environments.

[0094] The edge side is simulated and managed by the pSpace system, responsible for the lowest-level data acquisition and playback. The edge side samples hydraulic signals from seven racks (F1 to F7), including various parameters such as displacement, pressure, current, back pressure, and temperature. To verify system performance and model adaptability, pSpace provides a historical data playback function, capable of resending stored production data to the edge nodes at a fixed sampling period to simulate the data flow under real-world operating conditions. The edge side's acquisition cycle is 50ms, enabling the reproduction of on-site signal changes without affecting actual production, providing a stable data source for system debugging and model verification. The acquired multi-channel data is packaged by pSpace and transmitted over the network to the edge nodes for real-time computation and feature extraction.

[0095] The cloud, edge, and device layers interact with each other through standardized interfaces. Edge nodes periodically pull model updates and parameter configurations from the cloud, while simultaneously reporting local diagnostic results and system operating status to the cloud; the device side continuously sends sampling data in a streaming manner to achieve real-time data synchronization. Through this bidirectional communication mechanism between the upper and lower layers, the cloud can monitor the operating status of each edge node, while the edge nodes can automatically adjust their working modes or refresh model versions according to instructions from the cloud.

[0096] The overall system architecture is shown in Figure 4. The Django framework and FastAPI service are deployed on edge nodes, responsible for data processing and user interaction; the FlaskAPI service is deployed in the cloud for model training and performance management; and the pSpace module is deployed on the edge side to implement data collection and historical playback. Together, these three constitute a complete cloud-edge-device architecture. This architecture significantly reduces network latency caused by data uploading to the cloud by transferring a large amount of computation from the cloud to the local machine through the execution of the main data analysis and model inference on the edge side. The processed data results can be analyzed and displayed locally, greatly improving system response speed and real-time performance. In terms of network load and cost, this structure avoids frequent uploading of raw signals, reducing bandwidth consumption, lowering network transmission pressure, and reducing system maintenance and operation costs. Regarding security, the main data processing is completed locally at the edge, and sensitive data does not need to be stored in the cloud for extended periods, thus reducing the risk of data leakage and external intrusion. When communication between the cloud and the edge is abnormal or there are network fluctuations, the edge nodes can operate independently and continue to perform leak diagnosis without affecting real-time monitoring results due to communication interruptions, thereby ensuring system stability and continuity.

[0097] Furthermore, it should be noted that the database used in this embodiment is a MySQL database, mainly used to store training and testing data. To reduce network overhead during training and testing, this embodiment deploys the database along with the core training and testing code on a cloud server. The training and testing data are stored in two databases, `train` and `test`, respectively, within the MySQL database. Seven tables, F1 to F7, are created within these databases to store data for each of the seven racks. Simultaneously, oil leakage anomaly data during production at racks F1 to F7 is also stored in the cloud database as a record of abnormal operating conditions. When the real-time status monitoring process of the side mill diagnoses abnormal data, the corresponding processing function in Django pushes the abnormal data to the front-end page and also stores the abnormal data in the `log` database. A `diagnosis_results` table is created in the `log` database, storing the coil number, rack number, the number of times the three types of faults occurred during the coil rolling process, and the time when the coil rolling process ended. The time is used to facilitate subsequent queries of historical abnormal statuses.

[0098] This system uses the Django framework as the main framework to develop the "Hot Rolling Process Hydraulic System Oil Leakage Monitoring" system, serving as the web front-end for monitoring oil leakage in the hot rolling process hydraulic system. Simultaneously, the Flask framework is used to build the Flask API service application, separating the training and testing code from the Django framework. This allows the Django application to act as a backend for frontend (BFF), primarily handling user requests sent by clicking on the front-end page. It uses the requests library to call the Flask API service application's routing, forwarding user requests to the Flask microservices—the training and testing code—and encapsulating the training or testing results in JSON format before returning them to the front-end. Throughout this process, the Flask application, as an independent microservice, possesses the core training and testing code, only receiving requests from Django view functions and providing model training and testing capabilities. This model allows the diagnostic process to continue without stopping the Django service, while directly replacing the Flask microservice for algorithm upgrades. Figure 5 illustrates this process.

[0099] The completed system includes three functional pages: real-time mill status display, model training and testing, and data management. On the real-time mill status display page, users can monitor the mill's status in real time. When a mill malfunction occurs, the fault information is recorded and displayed on the front-end page. The data management page includes three sub-pages: data replacement / filling, data visualization, and data deletion. The data replacement / filling page allows users to replace or fill training or test sets; the data visualization page maps historical text data of the hot strip mill hydraulic system into graphs using the JavaScript ECharts visualization library; and the data deletion page allows users to selectively delete existing data. The model training and testing function can utilize data pre-stored on the data management page for training and testing. The front-end follows Django's MTV architecture, providing pages for the system homepage, real-time status, model training and testing, data management, and historical query. The real-time status page displays position, pressure, current, and back pressure curves as graphs, along with threshold lines and alarm indicators. When an anomaly is detected, a pop-up window appears on the front-end, triggering a back-end record. The system interface is shown in Figure 6.

[0100] To comprehensively evaluate the effectiveness and reliability of this invention, system verification was conducted based on historical operating data from a steel plant's 2150 hot strip mill production line. Regarding data partitioning, the dataset was strictly divided into training, validation, and test sets according to chronological order, ensuring the temporal consistency of data distribution and the fairness of the evaluation process.

[0101] During the model training and validation phases, the constructed CNN-LSTM-MultiHeadAttention fusion model performed exceptionally well. Specifically, the accuracy during training reached 0.9931, and the accuracy during validation was even higher at 0.9948, both achieving extremely high levels. Meanwhile, the training loss and validation loss were controlled at low levels of 0.0273 and 0.0215, respectively. This combination of high accuracy and low loss fully demonstrates the model's excellent learning ability and fitting performance, without exhibiting significant overfitting. Evaluation results on the independent test set further confirm the model's powerful performance. The CNN-LSTM-MultiHeadAttention fusion method achieved an accuracy of 0.9900 on the test set, demonstrating excellent generalization ability. Furthermore, the model's precision was 0.9903, recall was 0.9852, and F1-Score was 0.9877, all maintaining high levels of these key metrics. High precision indicates that the model has an extremely low false alarm rate when identifying oil leaks, while high recall demonstrates the model's excellent ability to detect real oil leak events. The balanced F1-Score comprehensively reflects the model's overall superior performance in oil leak monitoring tasks. These quantitative indicators collectively demonstrate the high reliability and practicality of the proposed solution for hydraulic system oil leak monitoring. Detailed test results are visualized in Figures 2 and 3.

[0102] Practical application tests under typical operating conditions show that the system can issue timely warnings after an oil leak occurs and accurately distinguish between different types such as "internal leakage," "servo valve leakage," and "back pressure side leakage," providing accurate fault diagnosis information for on-site maintenance personnel. To further verify the rationality of the model architecture design, this embodiment also conducted ablation experiments on the system. The experimental results show that the model accuracy drops significantly after removing the multi-head attention mechanism; similarly, removing the convolution module also leads to a significant performance reduction. This phenomenon strongly demonstrates that the three modules—CNN feature extraction, LSTM temporal modeling, and the attention mechanism—each perform their respective functions and work together to contribute to the overall performance improvement. Detailed comparison of ablation experiment results is shown in Table 2.

[0103] Table 2 Ablation Experiment Results

[0104]

[0105] In summary, this invention achieves hierarchical management of computing tasks and closed-loop management of information flow through a cloud-edge-device collaborative mechanism. This ensures both the real-time performance and stability of hydraulic system oil leakage detection, while also considering the efficiency of network resource utilization and system security. Furthermore, rigorous experimental verification and comprehensive performance evaluation demonstrate that the cloud-edge collaborative hot continuous rolling hydraulic system oil leakage monitoring system proposed in this invention exhibits excellent performance in terms of accuracy, real-time performance, and reliability, fully meeting the actual needs of industrial sites for hydraulic system oil leakage monitoring.

[0106] Second Embodiment

[0107] This embodiment provides an electronic device, as shown in FIG7. The electronic device includes a processor and a memory; wherein the processor and the memory can be connected via a communication bus; the memory stores at least one instruction, which is loaded and executed by the processor to implement the method of the first embodiment described above. Furthermore, the electronic device may also include a transceiver, and the processor and the transceiver can be connected via a communication bus, the transceiver being used to communicate with other devices.

[0108] The following is a detailed description of each component of this electronic device, with reference to Figure 7:

[0109] The processor is the control center of the electronic device. The electronic device may include multiple processors, each of which can be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The term "processor" can refer to a single processor or a collective term for multiple processing elements. For example, a processor can be one or more central processing units (CPUs), other general-purpose processors, application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), one or more field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor can perform various functions of the electronic device by running or executing software programs stored in memory and by calling data stored in memory.

[0110] In a specific implementation, as one example, the processor may include one or more CPUs, such as CPU0 and CPU1 shown in FIG7. Of course, this is only an illustrative example.

[0111] The memory is used to store the software program that executes the solution of the present invention, and the processor controls its execution. For specific implementation methods, please refer to the above method embodiments, which will not be repeated here.

[0112] Optionally, the memory may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory may be integrated with the processor or may exist independently and be coupled to the processor through the interface circuit of the electronic device (not shown in Figure 7). This embodiment of the invention does not specifically limit this.

[0113] The transceiver may include a receiver and a transmitter (not shown separately in Figure 7). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function. The transceiver may be integrated with the processor or may exist independently and be coupled to the processor through the interface circuit of the electronic device (not shown in Figure 7). This embodiment of the invention does not specifically limit this.

[0114] Furthermore, it should be noted that the structure of the electronic device shown in Figure 7 does not constitute a limitation on the device. Actual devices may include more or fewer components than shown, or combine certain components, or have different component arrangements. Moreover, the technical effects achieved by this electronic device when performing the method of the first embodiment described above can be referred to the technical effects described in the first embodiment; therefore, they will not be repeated here.

[0115] Fourth embodiment

[0116] This embodiment provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the method of the first embodiment described above. The computer-readable storage medium may be a ROM, random access memory, CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc. The instruction stored therein can be loaded and executed by a processor in a terminal.

[0117] Furthermore, it should be noted that the present invention can be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention can take the form of a completely or partially hardware embodiment, a completely or partially software embodiment, or an embodiment combining software and hardware aspects. Moreover, when implemented in software, embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any usable medium accessible to a computer or a data storage device such as a server or data center containing one or more sets of usable media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive (SSD).

[0118] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, create means for implementing the functions specified in one or more blocks of the flowchart illustrations and / or one or more blocks of the block diagrams.

[0119] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal equipment to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams, whereby the instructions that execute on the computer or other programmable terminal equipment provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.

[0120] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element. Furthermore, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone, where A and B can be singular or plural. Additionally, the character " / " in this text generally indicates an "or" relationship between the preceding and following objects, but it can also indicate an "AND / OR" relationship. Please refer to the context for specific interpretations. "At least one" refers to one or more items, while "more than" refers to two or more items. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or multiple items. For example, at least one of a, b, or c can be represented as: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0121] Furthermore, it is understood that in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0122] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0123] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of functional modules / units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual couplings or direct couplings or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms. Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, i.e., they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs. Additionally, the functional units in the various embodiments of this invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0124] If the method is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0125] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments of the present invention have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make several improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A cloud-edge-end coordinated hydraulic system for monitoring oil leakage in a hot continuous rolling process, characterized in that, The system adopts a cloud-edge-device architecture, including a cloud, an edge, and an end. The end is used to collect operating parameters of the hydraulic reduction system of the finishing mill in the hot continuous rolling process. The edge is used to preprocess the operating parameters collected by the end and input the preprocessed parameters into a detection model trained in the cloud. The detection model outputs detection results to detect whether an oil leakage fault has occurred in the hydraulic reduction system of the finishing mill in the hot continuous rolling process. The detection model integrates a convolutional neural network, a long short-term memory network, and a Transformer multi-head attention mechanism. The cloud is used for training and updating the detection model. The operating parameters include the pressure, position, and flow rate of the hydraulic cylinder back pressure chamber and working chamber of each stand in the hot continuous rolling process finishing mill, as well as the servo valve current of each stand's hydraulic system. The detection model includes a first feature extraction module, a second feature extraction module, and an attention module. The second feature extraction module includes a CNN network and an LSTM network. The attention module uses a T... Ransformer's multi-head attention mechanism; the detection model's processing of input data includes: firstly, dividing the input data into multiple samples using a sliding window; then, performing one-dimensional convolution, batch normalization, and max pooling operations on each sample using a first feature extraction module to extract first data features; then inputting the first data features into a second feature extraction module and an attention module respectively; the second feature extraction module first uses a CNN network to perform convolution, batch normalization, and max pooling operations on the first data features to extract second data features; then inputting the second data features into an LSTM network to extract third data features; the attention module uses multiple heads to focus on different aspects of the first data features and capture the global correlation between different features in the data to extract fourth data features; the third and fourth data features are fused to obtain fused features; the fused features are mapped to the classification space through a fully connected layer to distinguish between normal states and different types of oil leak states.

2. The oil leakage monitoring system for the hydraulic system of the hot continuous rolling process with cloud-edge-end coordination as described in claim 1, characterized in that, The preprocessing includes: standardizing the working parameters; and applying a windowing operation to the standardized working parameters.

3. The oil leakage monitoring system for the hydraulic system of the hot continuous rolling process with cloud-edge-end coordination as described in claim 1, characterized in that, The output of the detection model is the type of oil leakage in the hydraulic system during the hot rolling process; the types of oil leakage in the hydraulic system include: no leakage, leakage inside the hydraulic cylinder, leakage on the back pressure side, and leakage inside the servo valve.

4. The oil leakage monitoring system for the hydraulic system of the hot continuous rolling process with cloud-edge-end coordination as described in claim 1, characterized in that, The loss function of the detection model is a weighted combination of cross-entropy and L2 regularization.

5. The oil leakage monitoring system for the hydraulic system of the hot continuous rolling process with cloud-edge-end coordination as described in claim 1, characterized in that, The detection model employs an early stopping strategy and adaptive learning rate scheduling during training.

6. The oil leakage monitoring system for the hydraulic system of the hot continuous rolling process with cloud-edge-end coordination as described in claim 1, characterized in that, The attention module uses query-key interaction calculations with four independent attention heads to uncover implicit correlations between different time steps, and the attention weight matrix of the multi-head attention mechanism is normalized by the LayerNormalization layer; the dropout rate is set to 0.25 to prevent overfitting.

Images