Digital factory equipment operation state management system and method based on deep learning

Abstract

The invention discloses a digital factory equipment operation state management system and method based on deep learning, and belongs to the field of equipment management. The system comprises a data acquisition module, a data preprocessing module, a deep learning model construction module, an equipment state evaluation module, an abnormity early warning module, a maintenance suggestion generation module, a system integration module and a user interaction module. According to the method, the time sequence characteristics and the cross-modal information are fused through the multi-modal gating circulation unit, and the technologies of sliding window segmentation, PCA dimension reduction and the like are combined, so that real-time calculation and dynamic early warning of the equipment health index are realized. Therefore, the system can advance the downtime of the equipment fault, and the maintenance cost of the equipment is reduced. Therefore, the production efficiency can be indirectly improved. And compared with a traditional system, the system realizes multi-mode equipment operation state management, so that potential faults of the equipment can be revealed more comprehensively, and equipment maintenance personnel can know abnormities of the equipment more comprehensively.

Description

Technical Field

[0001] The present invention relates to equipment management, and particularly to a digital factory equipment operation status management system and method based on deep learning. Background Art

[0002] Traditional factory equipment management relies on manual inspections and regular maintenance, suffering from problems such as insufficient real-time performance, lagging fault response, and high maintenance costs. With the development of Industry 4.0, Internet of Things (IoT) and deep learning technologies have provided new ideas for equipment status monitoring: by collecting multi-dimensional data such as temperature, vibration, and current in real time through sensors, combined with edge computing and deep learning models, accurate prediction and proactive maintenance of equipment operation status can be achieved. However, existing systems mostly focus on single-modal data (such as only analyzing vibration signals), with limited cross-modal joint analysis capabilities and difficulty in comprehensively revealing potential equipment failures. Summary of the Invention

[0003] Object of the Invention: The object of the present invention is to provide a digital factory equipment operation status management system and method based on deep learning; to solve the problem that existing systems are difficult to comprehensively reveal potential equipment failures.

[0004] Technical Solution: A digital factory equipment operation status management system based on deep learning includes a data acquisition module, a data preprocessing module, a deep learning model construction module, an equipment status evaluation module, an anomaly warning module, a maintenance recommendation generation module, a system integration module, and a user interaction module;

[0005] The data acquisition module is used to collect equipment operation data in real time through sensors deployed in the equipment, including temperature, vibration, current, and environmental parameters;

[0006] The data preprocessing module is used to perform principal component analysis on the equipment operation data collected by the data acquisition module and retain the principal components with a cumulative contribution rate ≥ 95%;

[0007] The deep learning model construction module constructs the deep learning model using a multi-modal gated recurrent unit;

[0008] The equipment status evaluation module is used to evaluate the operation status of the equipment, obtain the equipment health index through the probability density function output by the deep learning model, and evaluate the equipment operation status based on this index;

[0009] The anomaly warning module determines whether the equipment has an anomaly based on the generated anomaly score, and generates a warning work order when the equipment has an anomaly and pushes it to the maintenance personnel's terminal device;

[0010] The maintenance recommendation generation module automatically generates a preventive maintenance plan according to the remaining service life of the equipment and synchronizes it to the ERP system;

[0011] System integration module, which integrates the system with the factory MES / SCADA system to achieve real-time data interaction between systems;

[0012] User interaction module, which enables human-computer interaction between the system and users, provides a 3D visualization dashboard, and real-time displays the equipment status heat map, fault trend prediction curve, and maintenance cost analysis.

[0013] Preferably, the data acquisition module uses an edge computing gateway to perform preliminary cleaning and format conversion on the original data, removing noise data and normalizing it to an interval.

[0014] Preferably, when the data preprocessing module performs principal component analysis on the cleaned data, there is:

[0015]

[0016] Among them, X is the original data matrix, U and V are the eigenvector matrices, ∑ is the eigenvalue diagonal matrix, and T is the window length.

[0017] Preferably, when constructing the deep learning model using a multi-modal gated recurrent unit:

[0018]

[0019] Among them, x t is the current input, h t-1 is the previous hidden state, z t is the gating signal, ⊙ is the element-wise product, W x and W h are the weight matrices, b n is the learnable parameter in batch normalization.

[0020] Preferably, the equipment status evaluation module calculates the equipment health index through the following formula:

[0021]

[0022] Among them, θ is the preset threshold, f(x) is the probability density function output by the model, and μ and σ 2 are the mean and variance respectively.

[0023] Preferably, the anomaly warning module generates an anomaly score based on a temporal convolutional network, specifically:

[0024]

[0025] Among them, Conv k is the k-th layer convolutional kernel, W k is the weight matrix, and K = 12 layers.

[0026] Preferably, the maintenance recommendation generation module is based on the prediction result of the remaining service life of the equipment, specifically:

[0027]

[0028] wherein, is the predicted health status of the i-th device, and N is the total number of similar devices.

[0029] Preferably, the user interaction module also supports custom report generation, exporting in PDF / Excel format, including equipment health index and energy consumption analysis index parameters.

[0030] A method for managing the operation status of digital factory equipment based on deep learning, including:

[0031] S1. Integrate the system into the factory's MES / SCADA system, enable real-time data interaction between the system and the factory's MES / SCADA system, and provide a man-machine interaction function at the same time;

[0032] S2. Real-time collect the temperature, vibration, current, and environmental parameters during the operation of the equipment through the data collection module, and hand over the collected data to the data preprocessing module for preprocessing;

[0033] S3. The data preprocessing module performs preliminary cleaning and format conversion on the equipment operation parameters collected by the data collection module, removes noise data and standardizes it to an interval;

[0034] S4. Use a multi-modal gated recurrent unit to construct the deep learning model construction module;

[0035] S5. Calculate the health index of the equipment through the probability density function output by the constructed deep learning model. When the health index < 0.7, trigger an early warning. When the health index < 0.4, start the emergency shutdown procedure;

[0036] S6. At the same time, generate an anomaly score through the anomaly warning module based on the temporal convolutional network. When the anomaly score > 0.8, generate a warning work order and push it to the maintenance personnel's terminal device;

[0037] S7. Then, according to the prediction result of the remaining service life of the equipment by the maintenance recommendation generation module, when the prediction result < 30 days, automatically generate a preventive maintenance plan and synchronize it to the ERP system.

[0038] Beneficial effects: By integrating temporal features and cross-modal information through a multi-modal gated recurrent unit, and combining technologies such as sliding window segmentation and PCA dimensionality reduction, the present invention realizes the real-time calculation and dynamic early warning of the equipment health index. Thereby, the system can advance the downtime of equipment failures and reduce the maintenance cost of equipment. Thereby, the production efficiency can be indirectly improved. And compared with traditional systems, this system realizes the multi-modal management of equipment operation states, thereby being able to more comprehensively reveal potential faults of equipment, and further enabling equipment maintenance personnel to more comprehensively understand the anomalies existing in the equipment and handle them in a timely manner, so that the equipment can eliminate anomalies in a timely manner, thereby improving production efficiency. Brief Description of the Drawings

[0039] Figure 1 It is a schematic diagram of the system flow of the present invention. Detailed Embodiments

[0040] To make the technical solutions of the present invention clearer, the following further describes the present invention in detail with reference to the accompanying drawings and specific embodiments.

[0041] As Figure 1 shown, the present invention discloses a digital factory equipment operation state management system and method based on deep learning, specifically including:

[0042] Data acquisition module: Real-time acquisition of equipment operation data through sensors, including temperature, vibration, current, environmental parameters, etc., and the data acquisition frequency ≥ 10Hz. And use an edge computing network to perform preliminary cleaning and format conversion on the original data, remove noise and standardize it to the interval.

[0043] Data preprocessing module: Perform principal component analysis (PCA) on the cleaned data:

[0044]

[0045] Among them, X is the original data matrix, U and V are the eigenvector matrices, ∑ is the eigenvalue diagonal matrix, and T is the window length. For time series data, use a sliding window segmentation, with a window length of 2048 sampling points and a step size of 512 sampling points.

[0046] Deep learning construction module: Use a multi-modal gated recurrent unit (MGRU):

[0047]

[0048] Among them, x t is the current input, h t-1 is the previous hidden state, z t is the gating signal, ⊙ is the element-wise product, W x and W h are weight matrices, bn They are learnable parameters in batch normalization, including a scaling factor and an offset factor. The model parameters are initialized using the Kaiming uniform distribution, the learning rate α = 1e-4, and the Adam optimizer is adopted.

[0049] Device status evaluation module: Calculate the device health index (DHI) through the probability density function (PDF) output by the deep learning model:

[0050]

[0051] where θ is a preset threshold, f(x) is the probability density function output by the model, and μ and σ 2 are the mean and variance respectively. When DHI < 0.7, a warning is triggered, and when DHI < 0.4, an emergency shutdown procedure is initiated.

[0052] Abnormal warning module: Generate an abnormal score based on the temporal convolutional network (TCN):

[0053]

[0054] where Conv k is the k-th layer convolutional kernel, W k is the weight matrix, and K = 12 layers. When the abnormal score A t > 0.8, a warning work order is generated and pushed to the maintenance personnel's terminal device.

[0055] Maintenance recommendation generation module: According to the prediction result of the remaining useful life (RUL) of the device:

[0056]

[0057] where is the predicted health status of the i-th device, and N is the total number of similar devices. When RUL(t) < 30 days, a preventive maintenance plan is automatically generated and synchronized to the ERP system.

[0058] System integration module: Real-time data interaction with the factory MES / SCADA system through the OPC UA protocol, with a latency ≤ 100 ms. Supports hybrid deployment of the cloud and the edge. The edge end processes latency-sensitive data, and the cloud is responsible for model iteration.

[0059] User interaction module: Provide a 3D visualization dashboard to display the device status heat map, fault trend prediction curve, and maintenance cost analysis in real time. Supports the generation of custom reports and can export in PDF / Excel format, including 12 indicators such as the device health index and energy consumption analysis.

[0060] Example 1: Injection molding machine temperature-vibration combined monitoring system

[0061] Data Acquisition and Preprocessing

[0062] Deploy sensors at the installation positions of the injection molding machine mold temperature controller (temperature sensor PT100, accuracy ±0.1°C) and the robot (vibration accelerometer ICP type, range ±50g), with a sampling frequency of 20Hz.

[0063] Use an edge computing gateway to clean the raw data, remove noise data (threshold set to ±3σ) and normalize it to the [0,1] interval.

[0064] Segment the time series data using a sliding window, with a window length of 2048 sampling points (corresponding to 10.24 seconds) and a step size of 512 sampling points (5.12 seconds).

[0065] Multi-modal Gated Recurrent Unit Model Construction

[0066] Model structure: The input layer is a 128-dimensional time series feature (temperature + vibration), the hidden layer uses MGRU units, and the output layer is a health status classification (normal / abnormal).

[0067] Calculation of the gating signal:

[0068]

[0069] where x t is the current input feature vector, h t-1 is the previous hidden state, z t is the update gate, r t is the reset gate, and σ is the Sigmoid activation function.

[0070] Parameter initialization: The weight matrix uses the Kaiming uniform distribution, the bias term is initialized to 0, the learning rate α = 1e−4, and the optimizer is Adam.

[0071] Calculation of the Equipment Health Index

[0072] Calculate the DHI through the probability density function output by the model:

[0073]

[0074] where θ = 0.95 is the preset threshold, f(x) is the probability density function output by the model, and μ and σ 2 are the mean and variance respectively. When DHI < 0.7, an alarm is triggered, and when DHI < 0.4, an emergency shutdown procedure is initiated.

[0075] Application Effect

[0076] After implementation in a certain auto parts factory, the equipment failure downtime was reduced by 42%, the maintenance cost was reduced by 28%, the temperature prediction error was < ±0.5°C, and the vibration amplitude prediction error was < ±2g.

[0077] Example 2: Motor Current-Temperature Multimodal Analysis System

[0078] Data Acquisition and Feature Extraction

[0079] Collect the three-phase current of the motor through the PLC (16-bit ADC accuracy, range 0-200A), and synchronously collect the winding temperature (K-type thermocouple, accuracy ±1°C).

[0080] Perform wavelet transform denoising on the current signal, and use Savitzky-Golay filtering for smoothing processing of the temperature signal.

[0081] Align the multimodal data, and control the timestamp error within ±100ms.

[0082] Deep Learning Model Construction

[0083] Adopt a 1D-CNN + Transformer hybrid architecture:

[0084] 1D-CNN layer: The convolution kernel size is 32×16, the stride is 8, the activation function is ReLU, and the number of output channels is 64;

[0085] Transformer layer: The number of heads is 8, the hidden layer dimension is 128, and the feed-forward network dimension is 512;

[0086] Output layer: The fully connected layer maps to the health status classification (normal / abnormal);

[0087] The temporal convolutional network generates an anomaly score:

[0088]

[0089] Among them, Conv k is the k-th layer convolution kernel, and W k is the weight matrix. When the anomaly score A t > 0.8, a warning work order is generated and pushed to the terminal device of the maintenance personnel.

[0090] Remaining Useful Life Prediction

[0091] Predict the motor life based on the convolutional LSTM model:

[0092]

[0093] Among them, It is the predicted health status of the i-th device, and N = 50 is the total number of similar devices. When RUL(t) < 30 days, a preventive maintenance plan is automatically generated and synchronized to the ERP system.

[0094] Application effect

[0095] After a certain steel enterprise applied it, the motor burnout accident rate decreased by 65%, the prediction maintenance accuracy rate increased to 89%, and the current harmonic analysis error < ±3%.

[0096] The above-described embodiments merely represent several implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the present invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the appended claims.

Claims

1. A digital factory equipment operation status management system based on deep learning, characterized in that, It includes a data acquisition module, a data preprocessing module, a deep learning model construction module, a device status evaluation module, an anomaly warning module, a maintenance suggestion generation module, a system integration module, and a user interaction module; The data acquisition module is used to collect the device operation data in real time through the sensors deployed in the device, including temperature, vibration, current, and environmental parameters; The data preprocessing module is used to perform principal component analysis on the device operation data collected by the data acquisition module and retain the principal components with a cumulative contribution rate ≥ 95%; The deep learning model construction module constructs the deep learning model using a multi-modal gated recurrent unit; The device status evaluation module is used to evaluate the operation status of the device, obtain the device health index through the probability density function output by the deep learning model, and evaluate the device operation status according to this index; The anomaly warning module determines whether the device has an anomaly based on the generated anomaly score, generates a warning work order when the device has an anomaly, and pushes it to the maintenance personnel's terminal device; The maintenance suggestion generation module automatically generates a preventive maintenance plan according to the remaining service life of the device and synchronizes it to the ERP system; The system integration module integrates this system with the factory MES / SCADA system to achieve real-time data interaction between systems; The user interaction module is used to enable human-computer interaction between the system and the user, provide a 3D visualization dashboard, and display the device status heat map, fault trend prediction curve, and maintenance cost analysis in real time.

2. The digital factory equipment operation status management system based on deep learning according to claim 1, characterized in that The data acquisition module uses an edge computing gateway to perform preliminary cleaning and format conversion on the original data, remove noise data, and standardize it to an interval.

3. The digital factory equipment operation status management system based on deep learning according to claim 2, characterized in that When the data preprocessing module performs principal component analysis on the cleaned data, there is: ； Where X is the original data matrix, U and V are the eigenvector matrices, ∑ is the eigenvalue diagonal matrix, and T is the window length.

4. The digital factory equipment operation status management system based on deep learning according to claim 1, wherein, When constructing the deep learning model using a multi-modal gated recurrent unit: ； where x t is the current input, h t-1 is the previous hidden state, z t is the gating signal, ⊙ is the element-wise product, W x and W h are weight matrices, b n is the learnable parameter in batch normalization.

5. The digital factory equipment operation status management system based on deep learning according to claim 1, wherein, The device status evaluation module calculates the device health index through the following formula: ； where θ is a preset threshold, f(x) is the probability density function output by the model, and μ and σ 2 are the mean and variance respectively.

6. The digital factory equipment operation status management system based on deep learning according to claim 1, characterized in that The anomaly warning module generates an anomaly score based on a temporal convolutional network, specifically: ； Among them, Conv k is the k-th layer convolutional kernel, and W k is the weight matrix, where K = 12 layers.

7. The digital factory equipment operation status management system based on deep learning according to claim 1, characterized in that The maintenance suggestion generation module is based on the prediction result of the remaining service life of the device, specifically: ； Among them, is the predicted health state of the i-th device, and N is the total number of devices of the same type.

8. The digital factory equipment operation status management system based on deep learning according to claim 1, wherein, The user interaction module also supports the generation of custom reports, which can be exported in PDF / Excel format and contain the device health index and energy consumption analysis index parameters.

9. A method for managing the operating status of digital factory equipment based on deep learning, characterized in that, This method is implemented based on the deep learning-based digital factory device operation status management system described in any one of claims 1 to 6, and specifically includes the following steps: S1. Integrate the system into the factory's MES / SCADA system, enable real-time data interaction between the system and the factory's MES / SCADA system, and at the same time provide a human-computer interaction function; S2. Collect the temperature, vibration, current, and environmental parameters during the device operation process through the data acquisition module, and hand over the collected data to the data preprocessing module for preprocessing; S3. The data preprocessing module performs preliminary cleaning and format conversion on the device operation parameters collected by the data acquisition module, removes noise data, and standardizes it to an interval; S4. Construct the deep learning model building module using a multi-modal gated recurrent unit; S5. Calculate the health index of the device through the probability density function output by the constructed deep learning model. Trigger an alarm when the health index < 0.7, and start an emergency shutdown procedure when the health index < 0.4; S6. At the same time, generate an anomaly score through the anomaly warning module based on a temporal convolutional network. When the anomaly score > 0.8, generate a warning work order and push it to the maintenance personnel's terminal device; S7. Then generate a preventive maintenance plan according to the prediction result of the remaining useful life of the device by the maintenance advice generation module. When the prediction result < 30 days, automatically generate a preventive maintenance plan and synchronize it to the ERP system.

Images