Online fault diagnosis method for petrochemical production equipment
By establishing edge models at edge nodes in petrochemical production devices and working in collaboration with cloud models to update model parameters in real time, the problem of online real-time fault diagnosis in the existing technology is solved, and efficient and real-time fault diagnosis is achieved, suitable for dynamically changing production environments.
Patent Information
- Application Number
- CN202411847011.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-16
- Publication Date
- 2025-03-28
AI Technical Summary
Existing deep learning methods are difficult to achieve online real-time fault diagnosis in real-time dynamic changing environments of petrochemical production devices, and are limited by storage, computing and network bandwidth.
By establishing the original edge model at edge nodes and working in concert with the cloud model, the edge model parameters are updated in real time, and edge cloud communication is only performed when new failures are diagnosed, reducing communication frequency and improving real-time performance.
It realizes online real-time fault diagnosis in petrochemical production equipment, improves diagnostic accuracy and efficiency, takes into account storage, computing and bandwidth resources, and is suitable for dynamically changing production environments.
Smart Images
Figure CN119719674A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of petrochemical equipment fault diagnosis, and in particular to an online fault diagnosis method for petrochemical production equipment. Background Art
[0002] In the petrochemical field, the fault diagnosis of petrochemical production equipment has a great impact on the entire production, and equipment failures often lead to immeasurable loss of life and property. Petrochemical production equipment operates all year round, so fault diagnosis requires continuous accumulation of fault knowledge at different time nodes to achieve a certain fault diagnosis recognition accuracy. At the same time, fault diagnosis also has high requirements for timeliness, so fault diagnosis methods also need to consider timeliness issues. In recent years, with the rapid development of technologies such as computers and artificial intelligence, data-driven methods represented by machine learning, especially deep learning, have become a current research hotspot. Data-driven fault diagnosis methods directly start from the collected status data, use deep neural network models to learn fault characteristics from the production equipment status data and realize automatic fault identification, which significantly improves the efficiency and intelligence level of fault diagnosis. However, existing deep learning methods such as deep belief neural network, convolutional neural network and long short-term memory neural network and other advanced algorithms need to collect all data and train the entire model at one time, which is not suitable for the real-time dynamic changes of petrochemical production equipment, and therefore is not suitable for faults such as petrochemical production equipment that require online real-time diagnosis; existing deep learning algorithm models are mostly deployed in cloud data centers (or remote servers) and other computing nodes with strong storage and computing capabilities, but this centralization and network bandwidth pressure will lead to delays and other problems, making it difficult to meet the needs of online diagnosis; some choose to be deployed on edge nodes, where edge computing is closer to the diagnosis object, and can provide storage, computing power and network bandwidth nearby, and can quickly diagnose needs, but due to storage and computing power limitations, edge nodes have certain pressure in dealing with the storage and computing of large amounts of data such as deep learning model algorithms. In summary, no single computing mode is the most ideal solution, and factors such as storage, computing power and network bandwidth are also factors that need to be considered when deploying fault diagnosis based on deep learning model algorithms. Therefore, petrochemical unit fault diagnosis urgently needs a method that can be applied to online dynamic advanced fault diagnosis technology and is not limited by storage, computing and network bandwidth. Summary of the invention
[0003] The purpose of the present invention is to overcome the deficiencies in the prior art and provide an online fault diagnosis method for petrochemical production equipment, which can cope with the dynamic changes in the working conditions of petrochemical production equipment by continuously learning new fault knowledge and improve the accuracy of fault diagnosis.
[0004] To achieve the above object, the present invention is implemented by adopting the following technical solutions:
[0005] The present invention provides a method for online fault diagnosis of petrochemical production equipment, comprising:
[0006] Obtaining the original cloud model and the original fault category library;
[0007] Sending the parameters of the original cloud model to the edge node to establish the original edge model;
[0008] Inputting the acquired real-time multimodal data of the petrochemical production equipment into the original edge model, updating the parameters of the original edge model and performing fault diagnosis on the real-time multimodal data to obtain an updated edge model and fault diagnosis results;
[0009] If the fault diagnosis result does not belong to the original fault category library, uploading the parameters of the updated edge model and the fault diagnosis result to the original cloud model, updating the original cloud model and adding the fault diagnosis result to the original fault category library to obtain an updated cloud model and an updated fault category library;
[0010] If the fault diagnosis result belongs to the original fault category library, the fault diagnosis result is directly output.
[0011] In the present invention, the edge-cloud model communicates only when the edge model diagnoses a new fault, reducing the communication frequency to improve the real-time performance of the edge-cloud model, and allowing the edge-cloud model to update the diagnosis online in real time when new data is received, thereby improving the real-time diagnosis efficiency of the model.
[0012] Optionally, the real-time multimodal data of the petrochemical production equipment includes at least temperature, humidity, displacement, pressure, noise, liquid level and vibration frequency.
[0013] Optionally, the training of the edge model includes:
[0014] Using the weight dynamic adjustment algorithm, a penalty term is added to the loss function of the edge model to train the edge model in real time. This increases the difficulty of changing these important parameters during the update process, so that the edge cloud model can better identify and classify both new and old data.
[0015] Optionally, the loss function is expressed as:
[0016] ;
[0017] In the formula, Indicates the current cloud model parameters; Represents the loss function for new data; Represents the parameters of the i-th edge model before updating; is the updated parameters of the i-th edge model; Indicates the importance of the old model; Indicates the importance of network parameters when building a cloud model.
[0018] Optionally, before inputting the acquired real-time multimodal data of the petrochemical production equipment into the original edge model, the method further includes preprocessing the real-time multimodal data of the petrochemical production equipment.
[0019] Optionally, also include:
[0020] Sending the updated parameters of the cloud model to the updated edge model;
[0021] Inputting the acquired real-time multimodal data of the petrochemical production equipment into the updated edge model, re-updating the parameters of the updated edge model and performing fault diagnosis on the real-time multimodal data to obtain a re-updated edge model and an updated fault diagnosis result;
[0022] Repeat until the preset stop condition is reached.
[0023] Optionally, obtaining the original cloud model and the original fault category library includes:
[0024] Obtain historical normal and abnormal multimodal data of petrochemical production equipment;
[0025] According to the historical normal and abnormal multimodal data of the petrochemical production equipment, a neural network is used to establish an original cloud model and an original fault category library.
[0026] Optionally, the cloud model communicates with the edge model according to a preset communication cycle.
[0027] Compared with the prior art, the present invention has the following beneficial effects:
[0028] The present invention copes with the dynamic changes in the operating conditions of petrochemical production equipment by continuously learning new fault knowledge, thereby improving the accuracy of fault diagnosis, and enabling the model to be updated online in real time when new data is received, which can greatly save the time of fault diagnosis while improving the accuracy, thereby improving the efficiency of equipment fault diagnosis; at the same time, the fault diagnosis model is trained through edge-cloud collaboration, and edge-cloud communication is selectively performed only when new fault categories are generated, which can take into account factors such as storage, computing, and bandwidth, thereby maximizing the timeliness of model training and improving equipment fault diagnosis and production efficiency; the weight dynamic adjustment algorithm adds a penalty term to the objective function to give higher weights to important model parameters, thereby increasing the difficulty of these important parameters being changed during the update process, so that the diagnostic model can better identify and classify both new and old data. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1Shown is a schematic diagram of the framework of an online fault diagnosis method for petrochemical production equipment in one embodiment of the present invention. DETAILED DESCRIPTION
[0030] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. The embodiments of the present invention and the technical features in the embodiments may be combined with each other unless there is a conflict.
[0031] The term "and / or" is only a description of the association relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " generally indicates that the related objects are in an "or" relationship.
[0032] Example 1
[0033] like Figure 1 As shown, this embodiment introduces a method for online fault diagnosis of petrochemical production equipment, which specifically includes the following steps:
[0034] According to the historical normal and abnormal multimodal data of petrochemical production equipment, the original cloud model and the original fault category library are established using the neural network model.
[0035] The parameters of the original cloud model are copied to the edge node to establish the edge model. The edge model uses the new data generated in real time for online training and update. When a new fault category is found in the edge model diagnosis, the edge model parameters at this time are uploaded to the cloud model. The cloud model is trained and updated according to the new edge model parameters, and the fault category library is updated at the same time. The cloud model sends the updated parameters to the edge model for the next round of online updates.
[0036] In this way, the edge and cloud models can respond to dynamic changes in equipment operating conditions by continuously learning new knowledge online through new data generated in real time, thereby improving the recognition accuracy of fault diagnosis, and enabling the model to be updated online in real time when new data is received. There is no need to collect all data and train the entire model at one time like traditional methods. Therefore, it can greatly save fault diagnosis time while improving accuracy and improve the efficiency of equipment fault diagnosis. At the same time, the fault diagnosis model is trained through edge-cloud collaboration, and edge-cloud communication is performed selectively (only when new categories are generated). This takes into account factors such as storage, computing, and bandwidth, thereby maximizing the timeliness of model training and improving equipment fault diagnosis and production efficiency.
[0037] Example 2
[0038] Based on Example 1, this example also has the following design:
[0039] Step 1: Based on the historical normal and abnormal multimodal data of petrochemical production equipment, a neural network is used to establish an original cloud model and an original fault category library to obtain prior knowledge.
[0040] Step 2: Send the parameters of the original cloud model to the edge node in the first round of the training cycle , build the original edge model , at this time i=1.
[0041] Step 3: Preprocess the real-time multimodal data of petrochemical production equipment and then input it into the original edge model , update the original edge model The parameters of the real-time multimodal data are used to perform fault diagnosis to obtain an updated edge model and fault diagnosis results; suppose a petrochemical production device includes m modal data, and the multimodal data comes from multiple sensors monitoring and the petrochemical equipment itself, including but not limited to temperature, humidity, displacement, pressure, noise, liquid level and vibration frequency;
[0042] When new tasks are generated At this moment, the number of sample categories is , then all sample sets are expressed as , then the task The sample category set is expressed as The edge model is trained and updated according to the new task, so that all and validation set The learned cloud model can be used on the test set The set of all category samples in Carry out accurate classification to achieve fault diagnosis tasks;
[0043] In order to keep the important parameters of the cloud model as much as possible, so that the new model can maintain high accuracy when learning old data and solve the problem of model forgetting, this embodiment adopts a weight dynamic adjustment algorithm to update the edge model, thereby effectively using the existing model knowledge and new data to update the model parameters;
[0044] The weight dynamic adjustment algorithm adds a penalty term to the objective function to give higher weights to important model parameters, thereby increasing the difficulty of changing these important parameters during the update process. This allows the diagnostic model to better identify and classify both new and old data. The specific implementation is that the weight dynamic adjustment algorithm is based on the loss function minimization optimization target to ensure model accuracy and reduce the probability of forgetting. The loss function minimization target is as follows:
[0045] ;
[0046] In the formula, Indicates the current cloud model parameters; Represents the loss function for new data; Represents the parameters of the i-th edge model before updating; is the updated parameters of the i-th edge model; Indicates the importance of the old model; Indicates the importance of network parameters when building a cloud model. The larger it is, the more important the corresponding parameter is.
[0047] If the fault diagnosis result does not belong to the original fault category library, the updated edge model Uploading the parameters and fault diagnosis results to the original cloud model, updating the original cloud model and adding the fault diagnosis results to the original fault category library to obtain an updated cloud model and an updated fault category library;
[0048] The updating of the cloud model includes the following steps:
[0049] set up For the troubleshooting task sequence, is the nth new fault diagnosis task, For historical diagnostic tasks, these task data are used to build cloud models .
[0050] Each task ,in Each task Contains the training set for the jth task , validation set and test set Each task Include ( ) device status categories, For the task The number of samples;
[0051] The updated cloud model parameters are sent to the edge model, and the edge model parameters are updated again, starting the next round of online update and diagnosis of the cloud model and edge model.
[0052] If the fault diagnosis result belongs to the original fault category library, the fault diagnosis result is directly output.
[0053] This embodiment allows the model to update the training model online when new data is received, thereby improving the efficiency of online real-time diagnosis. It communicates with the cloud model only when a new fault category is generated in the edge model, and has a certain communication cycle. It reduces the communication frequency compared to traditional methods and improves the real-time performance of the model.
[0054] That is, when this embodiment is applied, it responds to the dynamic changes in equipment operating conditions by continuously learning new knowledge online through the new data generated in real time by the edge model and the cloud model, realizes fault diagnosis of petrochemical production equipment, and improves the recognition accuracy of fault diagnosis; through advanced artificial intelligence processing technology combined with edge-cloud collaborative methods, the model is allowed to be continuously updated online upon receiving new data, without the need to collect all data and train the entire model at once as in traditional methods, and is therefore more suitable for remote online real-time diagnosis of petrochemical production equipment faults.
[0055] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.
[0056] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0057] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0058] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0059] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.
Claims
1. A method for online fault diagnosis of petrochemical production equipment, characterized in that: include: Obtaining the original cloud model and the original fault category library; Sending the parameters of the original cloud model to the edge node to establish the original edge model; Inputting the acquired real-time multimodal data of the petrochemical production equipment into the original edge model, updating the parameters of the original edge model and performing fault diagnosis on the real-time multimodal data to obtain an updated edge model and fault diagnosis results; If the fault diagnosis result does not belong to the original fault category library, uploading the parameters of the updated edge model and the fault diagnosis result to the original cloud model, updating the original cloud model and adding the fault diagnosis result to the original fault category library to obtain an updated cloud model and an updated fault category library; If the fault diagnosis result belongs to the original fault category library, the fault diagnosis result is directly output.
2. The method for online fault diagnosis of petrochemical production equipment according to claim 1, characterized in that: The real-time multimodal data of the petrochemical production equipment includes at least temperature, humidity, displacement, pressure, noise, liquid level and vibration frequency.
3. The method for online fault diagnosis of petrochemical production equipment according to claim 1, characterized in that: The training of the edge model includes: By using a dynamic weight adjustment algorithm, a penalty term is added to the loss function of the edge model to train the edge model in real time.
4. The method for online fault diagnosis of petrochemical production equipment according to claim 3 is characterized in that: The loss function is expressed as: ； In the formula, Indicates the current cloud model parameters; Represents the loss function for new data; Represents the parameters of the i-th edge model before updating; is the updated parameters of the i-th edge model; Indicates the importance of the old model; Indicates the importance of network parameters when building a cloud model.
5. The method for online fault diagnosis of petrochemical production equipment according to claim 1, characterized in that: Before inputting the acquired real-time multimodal data of the petrochemical production equipment into the original edge model, the method also includes preprocessing the real-time multimodal data of the petrochemical production equipment.
6. The method for online fault diagnosis of petrochemical production equipment according to claim 1, characterized in that: Also includes: Sending the updated parameters of the cloud model to the updated edge model; Inputting the acquired real-time multimodal data of the petrochemical production equipment into the updated edge model, re-updating the parameters of the updated edge model and performing fault diagnosis on the real-time multimodal data to obtain a re-updated edge model and an updated fault diagnosis result; Repeat until the preset stop condition is reached.
7. The method for online fault diagnosis of petrochemical production equipment according to claim 1, characterized in that: The acquisition of the original cloud model and the original fault category library includes: Obtain historical normal and abnormal multimodal data of petrochemical production equipment; According to the historical normal and abnormal multimodal data of the petrochemical production equipment, a neural network is used to establish an original cloud model and an original fault category library.
8. The method for online fault diagnosis of petrochemical production equipment according to claim 1, characterized in that: The cloud model communicates with the edge model according to a preset communication cycle.