Equipment operation state identification and prediction method and device based on improved W-GAN

Abstract

The invention designs an equipment operation state identification and prediction method and device based on an improved W-GAN, the method and device are applied to intelligent prediction of the operation state of key equipment of a welding machine, the function of predictive maintenance facing the operation state of the key equipment of the welding machine is added in a platform, and the device comprises a data acquisition module, a construction module and a predictive maintenance module. The data acquisition module is used for collecting equipment operation state data through equipment such as a monocular camera and a monitoring system, and the equipment operation state data comprise data in a normal state and data in an abnormal state; the construction module is used for generating rich and vivid data by using a WGAN model and simulating various conditions possibly occurring in the equipment; the predictive maintenance module is a butt welding machine equipment state judgment module, and is used for carrying out maintenance in advance, monitoring the operation state of equipment in real time based on a network model prediction result, and more accurately judging when a fault occurs. If a fault hidden danger is found, an alarm or repair command is automatically triggered. The method is high in reliability, safety and efficiency.

Description

Technical Field

[0001] This invention belongs to the field of intelligent prediction of the operating status of key equipment in welding machines, specifically a method and device for equipment operating status identification and prediction based on an improved W-GAN. Background Technology

[0002] Equipment operational status identification and prediction is a crucial application in the Industrial Internet of Things (IIoT). It aims to improve equipment reliability and stability, and reduce maintenance costs and downtime by monitoring and analyzing equipment operating conditions. Traditional rule-based and machine learning-based methods can achieve equipment status identification and prediction to some extent, but their effectiveness is limited. In recent years, methods based on Generative Adversarial Networks (GANs) have achieved significant results in image and speech recognition and have been introduced into equipment operational status identification and prediction. GANs train generators to generate data on possible future equipment states, helping to predict the state a device might be in at a future point in time. This is significant for equipment maintenance and scheduling, as it allows for the prediction of potential faults or anomalies, enabling appropriate preventative or corrective measures to mitigate the impact of equipment failures on production and operations. GANs can help improve the accuracy and reliability of equipment status identification and prediction, providing strong support for equipment health monitoring and fault prevention.

[0003] Current methods for identifying and predicting equipment operating status primarily rely on manual inspections. Inspectors conduct on-site checks at regular intervals, judging the welding machine's operating status by visual inspection. This extensive patrol approach is difficult to supervise and evaluate, and feedback from inspections is often delayed, severely impacting production efficiency and quality. If problems are not detected promptly, they may lead to abnormal welding machine operation and potential hazards. Summary of the Invention

[0004] Based on the aforementioned technical problems, a method and apparatus for identifying and predicting equipment operating status based on an improved W-GAN are proposed to address the issues of high equipment maintenance and operation costs, safety hazards for maintenance and construction personnel, and low operation and maintenance efficiency in identifying and predicting equipment operating status.

[0005] The technical solution adopted by the present invention to achieve the above objectives is as follows:

[0006] A method for identifying and predicting device operating status based on an improved W-GAN includes the following steps:

[0007] 1) Collect equipment operating status data as the raw dataset;

[0008] 2) Construct and train a device operation status prediction model based on the improved W-GAN;

[0009] 3) Determine whether the welding equipment is operating normally based on the model's prediction results.

[0010] Step 2) includes the following steps:

[0011] 2.1) Randomly divide the original dataset into a first dataset and a second dataset;

[0012] 2.2) The random noise obtained after processing the data of the first dataset through a fully connected DenseNet convolutional network is used as the input of the generator to generate fake samples for the device.

[0013] 2.3) The discriminator compares the data in the second dataset with the fake samples from the device operation, and adjusts the generator parameters by calculating the loss function.

[0014] By calculating the Wasserstein distance W(P) r P g The loss function of the discriminator is as follows:

[0015]

[0016] Among them, E (x，y) y is the expected value, γ is the joint distribution, x is the real sample, y is the fake sample of device operation generated by the generator, and ∏(P) r P g ) is the true sample distribution P r The device operation spurious sample distribution P generated by the generator g The set of all possible joint distributions, when combined, yields a lower bound on the expected value among all possible joint distributions.

[0017] Step 3) includes the following steps:

[0018] 3.1) A fully connected DenseNet convolutional network is used to process the normal device operating status dataset and the device operating status dataset to be detected, respectively, to obtain the first feature vector and the second feature vector;

[0019] 3.2) Calculate the cosine similarity score between the first eigenvector and the second eigenvector;

[0020] 3.3) Determine whether the cosine similarity exceeds the threshold. If it does not exceed the threshold, the operating status of the device to be detected is abnormal.

[0021] The cosine similarity score cosθ is:

[0022]

[0023] Where A represents the first feature vector, B represents the second feature vector, n represents the total number of feature vectors in the dataset, and i represents the i-th feature vector in the dataset.

[0024] A device for identifying and predicting equipment operating status based on an improved W-GAN, comprising:

[0025] The data acquisition module is used to collect equipment operating status data as the raw dataset;

[0026] The construction module is used to build and train a device operating status prediction model based on the improved W-GAN;

[0027] The predictive maintenance module is used to determine whether the welding equipment is operating normally based on the prediction results of the model.

[0028] The construction module includes:

[0029] The data preprocessing module is used to randomly divide the original dataset into a first dataset and a second dataset;

[0030] The generator module is used to take the random noise obtained after processing the data of the first dataset through a fully connected DenseNet convolutional network as input to generate fake samples for the device to run.

[0031] The discriminator module is used to compare the data from the second dataset with the fake samples running on the device, and to adjust the generator parameters by calculating the loss function.

[0032] The predictive maintenance module includes:

[0033] The convolutional network module is used to process the normal device operating status dataset and the device operating status dataset to be detected using a fully connected DenseNet convolutional network to obtain the first feature vector and the second feature vector.

[0034] The cosine similarity calculation module is used to calculate the cosine similarity score between the first feature vector and the second feature vector.

[0035] The operation status detection module is used to determine whether the cosine similarity exceeds the threshold. If it does not exceed the threshold, the operation status of the device under test is abnormal.

[0036] The present invention has the following beneficial effects and advantages:

[0037] This invention achieves more comprehensive and accurate status identification, enabling timely detection of potential equipment problems and prompt corrective action to prevent production interruptions and losses due to malfunctions. It helps managers prepare in advance, taking appropriate measures to avoid potential equipment failures or perform maintenance proactively, thereby reducing maintenance costs, extending equipment lifespan, and improving equipment reliability and stability. Furthermore, it enhances the intelligence level of the equipment. By introducing advanced artificial intelligence technology, the equipment possesses autonomous learning and adaptive capabilities, continuously optimizing status identification and prediction algorithms to adapt to changes in different environments and operating conditions. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating the process of identifying and predicting the operating status of the device according to the present invention.

[0039] Figure 2 This is a schematic diagram of the training process structure of the WGAN device operation status recognition and detection model of the present invention;

[0040] Figure 3 This is a schematic diagram of the predictive maintenance module structure of the present invention;

[0041] Figure 4 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0042] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0043] like Figure 4 As shown, a device for identifying and predicting equipment operating status based on an improved W-GAN includes: a data acquisition module, a construction module, and a predictive maintenance module. The data acquisition module collects equipment operating status data through devices such as monocular cameras and monitoring systems. The construction module trains an algorithm model based on the collected raw data. This model can generate rich and realistic new samples, and based on more diverse data samples, it can predict various possible conditions of the equipment. The predictive maintenance module can determine whether the welding machine is operating normally based on the prediction results of the network model, and thus determine whether maintenance is required.

[0044] The data acquisition module includes a monitoring system set up in the welding machine's working area and a camera installed in the working area, used to acquire data information on the welding machine's working status.

[0045] The building module includes:

[0046] The generator module receives random noise as input and synthesizes data related to the welding machine's operating status.

[0047] The discriminator module is used to evaluate the similarity between the generated data related to the welding machine's operating status and the real data, that is, the ability to distinguish between generated data and real data;

[0048] The loss function module measures the difference between the generator and the real image to guide the generator training process;

[0049] The adversarial training module improves the quality of generated welding machine operating status data by having the generator and discriminator compete and learn from each other.

[0050] An improved WGAN model is used to enhance training stability and generate sample quality through Wasserstein distance.

[0051] The improved W-GAN device operation status identification and prediction method and device module is characterized in that the feature vector processing module in the predictive maintenance module uses a fully connected DenseNet convolutional network to process the dataset to obtain corresponding feature vectors.

[0052] The improved W-GAN equipment operation status identification and prediction method and device module is characterized in that the predictive maintenance module mainly uses cosine similarity to judge the welding machine equipment status, and the cosine similarity formula is: Where A represents the first feature vector, B represents the second feature vector, and cosθ represents the cosine similarity score.

[0053] The aforementioned data augmentation module is based on Generative Adversarial Networks (GANs). GANs primarily consist of two network models: a generator model and a discriminator model. Random noise passes through the generator model to produce fake samples that resemble the real sample set. The discriminator model then compares these fake samples with the real samples, continuously adjusting the generator's parameters based on the differences between the fake and real samples to optimize the generated data and make it more realistic.

[0054] The Wasserstein distance calculation formula is as follows: Among them, ∏(P r P g ) is P r and P g The set of all possible joint distributions. For each possible joint distribution γ, a true sample x and a generated sample y, b can be obtained from sampling (x, y) ~ γ, and the distance ||xy|| between these samples can be calculated. Therefore, the expected value E of the distance between the sampled pairs under the joint distribution γ can be calculated. (x，y) ~γ[||xy||].

[0055] Example

[0056] A device based on an improved W-GAN, its workflow diagram is as follows: Figure 1 As shown, it includes: a data acquisition module, a construction module, and a predictive maintenance module. The data acquisition module collects equipment operating status data through devices such as monocular cameras and monitoring systems. Based on the collected raw data, a WGAN equipment operating status recognition and detection model is constructed. This model can generate rich and realistic new samples. The predictive maintenance module, based on even more diverse data samples, uses convolutional neural networks to predict various possible conditions of the equipment, determining whether the welding machine is operating normally and whether maintenance is required.

[0057] The WGAN model in the construction module mainly includes modules such as generators and discriminators; for example... Figure 2 As shown, the first dataset can simulate the random noise input of the original WGAN model generator to train the generator G, and can generate fake images F; the second dataset can simulate the labels of the original WGAN model discriminator D, and can be input into the discriminator D together with the fake images F to train the discriminator D, and can return the discrimination results to the generator G to update the generator G, thereby iteratively training the discriminator D and the generator G to obtain the WGAN device state recognition and prediction model.

[0058] The predictive maintenance module operates as follows: Figure 3 As shown, data in normal operating condition is processed into a first feature vector, and data to be predicted is processed into a second feature vector. The cosine similarity score between the first feature vector and the second feature vector is calculated, and it is determined whether the cosine similarity exceeds a set threshold. If not, the operating condition of the device to be detected is abnormal.

[0059] The specific steps for equipment operating status identification and prediction are as follows:

[0060] S1 Data Acquisition: The data acquisition module collects equipment operating status data through devices such as monocular cameras and monitoring systems.

[0061] S2 Data Preprocessing: The original dataset is divided into two parts, namely the first dataset and the second dataset, according to the device requirements.

[0062] S3 Network Model Construction: A network model for WGAN device operation status identification and prediction is constructed based on two datasets. Specifically, an original WGAN model is first constructed, which includes a generator and a discriminator. The first and second datasets are input into the original WGAN model to train it and obtain the WGAN device operation status identification and prediction model. The first dataset simulates random noise from the generator and is input into the generator of the original WGAN model to train the generator and generate fake images. The second dataset simulates the labels of the discriminator of the original WGAN model and is input into the discriminator of the original WGAN model along with the fake images to train the discriminator. The discrimination results are returned to the generator to update the generator, thereby iteratively training the discriminator and the generator to obtain the WGAN device operation status identification and prediction model.

[0063] The S4 device operating status determination method uses a fully connected DenseNet convolutional network to process the normal device operating status dataset and the device operating status dataset to be detected, respectively, to obtain corresponding first feature vectors and second feature vectors. The cosine similarity score between the first feature vector and the second feature vector is calculated, and it is determined whether the cosine similarity exceeds a set threshold. If not, the operating status of the device to be detected is abnormal.

[0064] This invention discloses a method and apparatus for identifying and predicting equipment operating status based on an improved W-GAN. Through a data acquisition module, a construction module, and a predictive maintenance module, it establishes a deep learning-based method and equipment for identifying and predicting equipment operating status, enabling real-time monitoring of welding machine equipment operating status. This invention addresses the problem of low efficiency in manual inspection of welding machine equipment, and also overcomes the safety deficiencies and inability to respond promptly to faults associated with manual inspection. This invention has the advantages of high practicality and low cost.

[0065] The above description is a preferred embodiment of the invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principles of the invention, and these improvements and modifications should be considered within the scope of protection of the invention.

Claims

1. A method for identifying and predicting device operating status based on an improved W-GAN, characterized in that, Includes the following steps: 1) Collect equipment operating status data as the raw dataset; 2) Construct and train a device operation status prediction model based on the improved W-GAN; 3) Determine whether the welding equipment is operating normally based on the model's prediction results.

2. The method for identifying and predicting device operating status based on an improved W-GAN according to claim 1, characterized in that, Step 2) includes the following steps: 2.1) Randomly divide the original dataset into a first dataset and a second dataset; 2.2) The random noise obtained after processing the data of the first dataset through a fully connected DenseNet convolutional network is used as the input of the generator to generate fake samples for the device. 2.3) The discriminator compares the data in the second dataset with the fake samples from the device operation, and adjusts the generator parameters by calculating the loss function.

3. The method for identifying and predicting device operating status based on an improved W-GAN according to claim 2, characterized in that, By calculating the Wasserstein distance W(P) r ,P g The loss function of the discriminator is as follows: Among them, E (x,y) y is the expected value, γ is the joint distribution, x is the real sample, y is the fake sample of device operation generated by the generator, and ∏(P) r ,P g ) is the true sample distribution P r The device operation spurious sample distribution P generated by the generator g The set of all possible joint distributions, when combined, yields a lower bound on the expected value among all possible joint distributions.

4. The method for identifying and predicting device operating status based on an improved W-GAN according to claim 1, characterized in that, Step 3) includes the following steps: 3.1) A fully connected DenseNet convolutional network is used to process the normal device operating status dataset and the device operating status dataset to be detected, respectively, to obtain the first feature vector and the second feature vector; 3.2) Calculate the cosine similarity score between the first eigenvector and the second eigenvector; 3.3) Determine whether the cosine similarity exceeds the threshold. If it does not exceed the threshold, the operating status of the device to be detected is abnormal.

5. The method for identifying and predicting equipment operating status based on an improved W-GAN according to claim 4, characterized in that, The cosine similarity score cosθ is: Where A represents the first feature vector, B represents the second feature vector, n represents the total number of feature vectors in the dataset, and i represents the i-th feature vector in the dataset.

6. A device for identifying and predicting equipment operating status based on an improved W-GAN, characterized in that, include: The data acquisition module is used to collect equipment operating status data as the raw dataset; The construction module is used to build and train a device operating status prediction model based on the improved W-GAN; The predictive maintenance module is used to determine whether the welding equipment is operating normally based on the prediction results of the model.

7. The device for identifying and predicting equipment operating status based on an improved W-GAN according to claim 6, characterized in that, The construction module includes: The data preprocessing module is used to randomly divide the original dataset into a first dataset and a second dataset; The generator module is used to take the random noise obtained after processing the data of the first dataset through a fully connected DenseNet convolutional network as input to generate fake samples for the device to run. The discriminator module is used to compare the data from the second dataset with the fake samples running on the device, and to adjust the generator parameters by calculating the loss function.

8. The device for identifying and predicting equipment operating status based on an improved W-GAN according to claim 6, characterized in that, The predictive maintenance module includes: The convolutional network module is used to process the normal device operating status dataset and the device operating status dataset to be detected using a fully connected DenseNet convolutional network to obtain the first feature vector and the second feature vector. The cosine similarity calculation module is used to calculate the cosine similarity score between the first feature vector and the second feature vector. The operation status detection module is used to determine whether the cosine similarity exceeds the threshold. If it does not exceed the threshold, the operation status of the device under test is abnormal.

Images