Deep Learning-Based Multi-Valve Fault Diagnosis Method and System for Air Compressors

Abstract

The present invention provides a method and system for diagnosing multi-valve faults of an air compressor based on deep learning. Each valve of the air compressor is connected to a computer, and the computer collects the current working parameters of the valve and inputs the current working parameters of the valve into the control center. The current fault type is judged by means of deep learning, and the operation process of the air compressor can be monitored in real time. The working parameters of each valve of the air compressor under normal working conditions are obtained by means of deep learning, and the current fault type of the air compressor can be judged according to different parameters. When a fault occurs in the air compressor, it can be discovered in time and early warning is given to prevent major problems from occurring due to lack of supervision during the operation of the air compressor, which affects production and even endangers safety, causing potential safety hazards and economic losses to coal mining work, improving the safety and reliability of the operation process of the air compressor, and solving the problem of diagnosing multi-valve faults of the air compressor.

Description

Technical Field

[0001] The present invention belongs to the technical field of reliability diagnosis, and particularly relates to a multi-valve fault diagnosis method for air compressors based on deep learning. Background Art

[0002] Currently, in the coal industry, as a very important air flow supply device, the reliability of the air compressor during operation is very important for the overall operation of the system. Each air compressor is equipped with multiple valves, and these valves may malfunction during operation, affecting the normal operation of the air compressor, causing production stagnation, reducing production efficiency, and resulting in economic losses.

[0003] The air compressor itself has several different valves. In the prior art, it is only possible to monitor the abnormal operation of the current air compressor, and it is difficult to determine whether there are valve failures in the air compressor, as well as the specific types of failures and the specific locations of the faulty valves. As a result, the phenomenon of the air compressor operating with faults is relatively serious. When the air compressor fails to load or unload, it will cause excessive pressure in the air compressor and there may also be leakage problems, which will seriously affect the production safety of the air compressor. Summary of the Invention

[0004] Based on the above problems, the present invention proposes a multi-valve fault diagnosis method for air compressors based on deep learning, including:

[0005] Step 1: Collect the operating state parameters of each valve and the corresponding fault types under different operating states of the air compressor to construct a sample set; specifically:

[0006] Step 1.1: Start the air compressor and collect the operating state parameters of all valves when the air compressor is operating normally.

[0007] Step 1.2: Adjust the operating state of the air compressor and collect the operating state parameters of each valve on the air compressor again. Continuously adjust the operating state of the air compressor and collect the operating state parameters of each valve under different operating states.

[0008] Step 1.3: Select some valves to close and collect the operating state parameters of each valve on the current air compressor; randomly select several valves to shut down and continuously collect the operating state parameters of each valve on the air compressor.

[0009] Step 1.4: Increase the working load of the air compressor until the air compressor reaches an overload state, and collect the operating state parameters of each valve under the overload state of the air compressor. Adjust the overload type and fault state of the air compressor and continuously collect the operating state parameters of each valve under the overload condition.

[0010] Step 1.5: Adjust the air compressor cooling system to reduce the working power of the air compressor cooling system. Collect the operating state parameters of each valve under the condition of overheating of the air compressor. Continuously adjust the working power of the air compressor cooling system to make the air compressor cooling system in different overheating ranges, and continuously collect the operating state parameters of each valve at the current working state of the air compressor.

[0011] Step 1.6: Replace some worn parts. For different fault phenomena such as insufficient air supply, excessive vibration, and excessive noise in the air compressor, collect the operating state parameters of each valve of the air compressor under different fault types.

[0012] Step 1.7: According to the valve operating state parameters collected in Steps 1.1 to 1.6 and the corresponding operating state of the air compressor, establish sample data including the working state of the air compressor, the fault type, and the corresponding valve operating state parameters to form a sample set.

[0013] Step 2: Build a fault monitoring model based on deep learning and train the model using the sample set. Specifically described as:

[0014] Step 2.1: Input the operating state parameters of each valve of the air compressor when a fault occurs, and implement the learning process of the deep learning model according to the operating state parameters under valve faults.

[0015] Step 2.2: Select the operating state parameters of a single valve fault of the air compressor and input them into the deep learning model. Replace the faulty valve in turn to ensure that the operating state parameters when all valves fail are input into the deep learning model, and the deep learning model can complete the recognition, and perform the single-layer training process of the deep learning model.

[0016] Step 2.3: Select the operating state parameters of any number and any position of valves of the air compressor when a fault occurs and input them into the deep learning model. Select multiple groups of different valves and different operating state parameters and input them into the deep learning model, and the deep learning model can complete the recognition, and perform the multi-layer training process of the deep learning model.

[0017] Step 2.4: Continue the training. Collect the operating state parameters of the valves during the actual operation of the air compressor, adjust the working state of the air compressor valves, use the deep learning model trained in Step 2.3 to identify and collect the current operating state parameters and their corresponding fault types, and perform the reinforcement training process of deep learning. After the reinforcement training, a fault detection model is obtained.

[0018] Step 3: Use the trained fault monitoring model to monitor the current working state of the air compressor.

[0019] For each air compressor, collect the operating status parameters of each valve on the air compressor and the corresponding operating status of the air compressor, and construct a sample set for training a deep learning model as the fault monitoring model of the air compressor.

[0020] A multi-valve fault diagnosis system for air compressors based on deep learning. Multiple valves are installed on each air compressor. The system includes a computer and a control center. The valves are electrically connected to the computer, and the computer is wirelessly connected to the control center.

[0021] The computer is used to collect the operating status parameters of the valves on each air compressor and transmit them to the control center.

[0022] The control center is used to construct a fault monitoring model based on a deep learning model.

[0023] The beneficial effects of the present invention are:

[0024] The present invention proposes a multi-valve fault diagnosis method and system for air compressors based on deep learning. The operating parameter characteristics of each valve are stored in the system computer. By using the one-to-one correspondence between the valves and the computer, it can ensure that the parameter information of each valve is independently transmitted without interference affecting the information transmission. On the other hand, in the form of one-to-one correspondence between the valves and the computer, when the system computer receives computer data, it can judge the current valve position according to the data situation, saving data processing time, reducing the interference in the data transmission process, improving the reliability of the data transmission process, and preventing the influence of noise in the data processing process on data reliability. The system computer judges the current working condition of the air compressor according to the received signal situation and the data information. If the air compressor fails, it will report the fault type and the position of the faulty valve to the technical center, avoiding potential safety hazards caused by insufficient monitoring, and creating a new idea for the design of a multi-valve fault diagnosis system for air compressors based on deep learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a schematic diagram of the multi-valve fault diagnosis system for air compressors based on deep learning in the present invention.

[0026] Figure 2 It is a schematic diagram of the training of the deep learning model in the present invention. Among them, Figures (a) to (d) are all schematic diagrams randomly selecting some faulty valves and some normal valves. Here, it is only a random abnormal selection for demonstrating the training process. In the actual training process, the valve fault situation and the number of training times are set to be much greater than 4 times.

[0027] Figure 3 It is a schematic diagram of the multi-valve fault diagnosis method for air compressors based on deep learning in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0028] The invention will be further described below in conjunction with the accompanying drawings and specific embodiments. This embodiment provides a method and system for diagnosing multi-valve faults of an air compressor based on deep learning, and constructs a self-training model for fault detection and diagnosis when multiple valves of the air compressor are loose, ensuring that the air compressor can automatically and effectively control and operate efficiently under standard pressure indicators.

[0029] Based on the deep learning model, this embodiment judges the valve fault type online. The construction process of the deep learning model is as follows:

[0030] 1) Establishment of the data set. Select a large number of operating state parameters of the air compressor valves with the same proportion of various fault states, and establish a data set of the operating state parameters of the air compressor valves. The data set includes various parameters under normal working conditions of the air compressor valves and various parameters of fault conditions.

[0031] 2) Training and testing of the model. Construct two data sets, one of which is the training set and the other is the test set. Input the operating state parameters of the air compressor valves, where any number of the operating state parameters of the valves are in the fault state and any other number of the operating state parameters of the valves are in the normal state. A large amount of such data is included in the training set and all are input. Then, input the data in the test set. The data in the training set and the test set do not overlap, that is, to check the recognition accuracy of the current model, and the data in the test set can also be input into the model to expand the data set included in the model and further strengthen the data set.

[0032] 3) Evaluation of the trained model. Input the operating state parameters of the air compressor, record the set of operating state parameters with errors, and improve and strengthen the model according to the data with errors. After the strengthening work is completed, select the operating state parameters for input again, use the trained model for recognition, record the current recognition result and the input parameters, obtain the number of correct recognitions, and calculate the recognition accuracy.

[0033] As Figure 3 shown, a method for diagnosing multi-valve faults of an air compressor based on deep learning proposed by the present invention mainly includes the collection of working parameters, storing the working parameters in a database for training a deep learning model, online judging the fault type through the constructed deep learning model, and feeding back the working parameters in the case of faults to the control center to provide technical support for later maintenance work. The specific steps are as follows:

[0034] Step 1: Collect the operating state parameters and corresponding fault types of each valve of the air compressor under different working states, and construct a sample set; specifically described as:

[0035] Step 1.1: Start the air compressor and collect the operating state parameters of all valves when the air compressor is running normally;

[0036] Step 1.2: Adjust the operating state of the air compressor, and collect the operating state parameters of each valve on the air compressor again. Continuously adjust the operating state of the air compressor, and collect the operating state parameters of each valve on the air compressor under different operating states;

[0037] Step 1.3: Select some valves to close, and collect the operating state parameters of each valve on the current air compressor; Randomly select several valves to shut down, and continuously collect the operating state parameters of each valve on the air compressor;

[0038] Step 1.4: Increase the working load of the air compressor until the air compressor reaches the overload state, collect the operating state parameters of each valve under the overload state of the air compressor, adjust the overload type and fault state of the air compressor, and continuously collect the operating state parameters of each valve under the overload condition of the air compressor;

[0039] Step 1.5: Adjust the cooling system of the air compressor, reduce the working power of the cooling system of the air compressor, collect the operating state parameters of each valve under the overheating condition of the air compressor, continuously adjust the working power of the cooling system of the air compressor, so that the cooling system of the air compressor is in different overheating ranges, and continuously collect the operating state parameters of each valve under the current working state of the air compressor;

[0040] Step 1.6: Replace some worn parts, and collect the operating state parameters of each valve of the air compressor under different fault types for different fault phenomena such as insufficient air supply, excessive vibration, and excessive noise of the air compressor;

[0041] Step 1.7: According to the valve operating state parameters collected in Step 1.1 to Step 1.6 and the corresponding operating state of the air compressor, establish sample data including the operating state of the air compressor, the fault type and the corresponding valve operating state parameters, and form a sample set;

[0042] Step 2: Construct a fault monitoring model based on deep learning, and use the sample set to train the model, as Figure 2 described; specifically expressed as:

[0043] Step 2.1: Input the operating state parameters of each valve of the air compressor in the case of failure, and store the operating state parameters under valve failure in the computer, which is the learning process of the deep learning model;

[0044] Step 2.2: Select the operating state parameters of a single valve failure of the air compressor and input them into the deep learning model. Replace the failed valve in turn to ensure that the operating state parameters of all valves in the case of failure are input into the deep learning model, and the fault monitoring model can complete the identification; that is, the single-layer training process of the deep learning model;

[0045] Step 2.3: Select the operating state parameters of valves at any number and any position of the air compressor with faults and substitute them into the deep learning model. Select multiple groups of different valves with different operating state parameter faults and substitute them into the deep learning model, and the deep learning model can complete the recognition; that is, the multi-layer training process of the deep learning model.

[0046] Step 2.4: Continue the training, collect the operating state parameters of the valves during the actual operation of the air compressor, manually adjust the working state of the air compressor valves, and use the deep learning model trained in Step 2.3 to identify and collect the current operating state parameters and their corresponding fault types, that is, the reinforcement training process of the deep learning. After the reinforcement training, a fault detection model is obtained.

[0047] Step 3: Use the trained fault monitoring model to monitor the current working state of the air compressor.

[0048] For each air compressor, collect the operating state parameters of each valve on the air compressor and the corresponding operating state of the air compressor, and construct a sample set for training the deep learning model as the fault monitoring model of the air compressor.

[0049] A multi-valve fault diagnosis system for air compressors based on deep learning. There are multiple valves installed on each air compressor. The system includes a computer and a control center. The valves are electrically connected to the computer, and the computer is wirelessly connected to the control center. The computer can collect and feedback the corresponding air compressor valve parameters to the system computer (i.e., the control center) in real time. The system computer can grasp the operating conditions of the air compressor in real time. Once a problem occurs, it will dispatch staff in time for handling.

[0050] The computer is used to collect the operating state parameters of the valves on each air compressor and transmit them to the control center.

[0051] The control center is used to construct a fault monitoring model based on the deep learning model. Receive the valve working parameters fed back by the computer, judge the current fault situation according to the parameters, and handle it in time once a problem occurs to ensure the reliability of the overall working process of the air compressor, prevent major faults from occurring due to lack of monitoring of the air compressor, which may affect the production progress and even endanger production safety, and improve the reliability and safety of the production process underground in coal mines.

[0052] Such as Figure 1As shown, the system includes a setting feedback system, which consists of multiple computers. The number of computers is the same as the number of air compressor valves, and the air compressor valves correspond to the computers one by one. Each computer only collects the working parameters of the air compressor valve it corresponds to, and feeds back the current parameter situation of the air compressor valve to the system computer. The system computer can judge the type of fault occurring in the current air compressor according to different parameters, and feed back the current fault type of the air compressor to the computer, completing the fault diagnosis of the air compressor to ensure that the air compressor can operate stably in actual work and prevent the air compressor from malfunctioning and affecting production. When a valve malfunctions, it often shows as the valve being loose, but the looseness is only the appearance of the fault. The manifestation forms and parameter situations of different valves malfunctioning are different. At this time, the computer connected to the valve feeds the parameters back to the system computer. The system computer can determine the type of fault occurring in the air compressor according to the current working parameters, and the system computer feeds back the fault situation to the technical center. The technical center will dispatch technicians to repair the air compressor to ensure the reliable operation of the air compressor.

[0053] For the multi-valve fault diagnosis system of the air compressor based on deep learning, a deep learning model is constructed. By using the method of deep learning, a fault diagnosis and self-training model is established. Any proportion of valves in different valves malfunction, and their parameters are collected through calculation and brought into the system computer. A database is established in the system computer, and training and data statistics are carried out repeatedly to ensure that the normal operation data of each valve and the operation data under various fault conditions are included in the system computer. In this way, a training model is formed. The training model covers the parameter data of each valve under normal working conditions and fault conditions. According to the above parameter data, the current system fault type can be quickly and timely checked. If a fault problem occurs, it will be promptly reported to the technical center. The technical center will dispatch technicians to the site for maintenance to prevent serious faults of the air compressor caused by insufficient monitoring, which will not only cause economic impacts, but may also induce safety accidents in severe cases, affecting the safety of the production process.

[0054] The computer is used to collect the operating parameters of the valves. Since there are multiple valves in the air compressor, the number of computers is set corresponding to the number of valves. Each computer only reads the operating parameters of the valve it corresponds to and feeds them back to the system computer. That is, the parameters of Valve A are collected by Computer A, and Computer A only collects the parameters of Valve A and will not collect the parameters of other valves. The operating parameters of Valve A are also only transmitted to Computer A and will not be transmitted to other computers, so as not to interfere with the data collection of other parts. After receiving the valve operating parameters fed back by the computer, the system computer will enter the corresponding deep learning model of the valve according to the current specific computer, and judge the current working condition according to the data of its corresponding deep learning model. If a fault occurs, it will judge the fault type according to the operating parameters, and judge the valve position according to the computer fed back to the system computer. The system computer will feed back the fault information and valve position information to the technical center, and the technical center will dispatch technicians for maintenance according to the fault type and valve position. This ensures the real-time monitoring of the working condition of the air compressor, quickly feedbacks problems when they occur, prevents safety accidents caused by the operation of the air compressor under fault conditions, and affects production safety; it can also avoid the lack of monitoring during the operation of the air compressor, failure to repair problems in time, resulting in major faults of the air compressor, causing the production to be unable to proceed normally and causing economic losses. The use of a multi-valve fault diagnosis system for air compressors based on deep learning can monitor the operation of the air compressor in real time and quickly, prevent air compressor failures caused by insufficient monitoring, which may cause more serious economic impacts and endanger production safety, and improve the safety and reliability of the air compressor during operation.

[0055] The present invention realizes a multi-valve fault diagnosis system for air compressors based on deep learning. Its system computer can read the valve working parameters collected by the computer, judge the current working condition according to the valve parameters, and if a fault occurs, it will feedback the judgment result to the technical center, and the technical center staff will quickly go to the site for maintenance. The present invention uses deep learning to process valve parameters and determine the parameter range under reasonable working conditions of the valve and the fault types corresponding to different parameters. Taking different types of faults as inputs, different valves of the air compressor are randomly selected. When a fault occurs in a different valve, the valve parameters of the air compressor under the fault condition are recorded. Repeat the above deep learning training process, randomly select different valves, and record them with different fault type parameters, so as to determine the valve parameters under normal working conditions and different types of fault conditions of the air compressor. According to the feedback information of different computers, judge the specific valve and fault type with a fault at present, and determine the air compressor fault type with this parameter, thus completing the diagnosis work of multi-valve faults of the air compressor and improving the reliability of the system operation.

Claims

1. A fault diagnosis method for multiple valves of an air compressor based on deep learning, characterized in that, Including: Step 1: Collect the operation state parameters of each valve under different working conditions of the air compressor and the corresponding fault types, and construct a sample set; Step 2: Construct a fault monitoring model based on deep learning, and use the sample set to train the model; Step 3: Use the trained fault monitoring model to monitor the current working state of the air compressor; The specific description of Step 1 is as follows: Step 1.1: Start the air compressor and collect the operation state parameters of all valves when the air compressor is running normally; Step 1.2: Adjust the working state of the air compressor, and collect the operation state parameters of each valve on the air compressor again. Continuously adjust the working state of the air compressor, and collect the operation state parameters of each valve on the air compressor under different working conditions; Step 1.3: Select some valves to close, and collect the operation state parameters of each valve on the current air compressor; Randomly select several valves to shut down, and continuously collect the operation state parameters of each valve on the air compressor; Step 1.4: Increase the working load of the air compressor until the air compressor reaches the overload state, and collect the operation state parameters of each valve under the overload state of the air compressor. Adjust the overload type and fault state of the air compressor, and continuously collect the operation state parameters of each valve under the overload condition of the air compressor; Step 1.5: Adjust the cooling system of the air compressor, reduce the working power of the cooling system of the air compressor, and collect the operation state parameters of each valve under the overheat condition of the air compressor. Continuously adjust the working power of the cooling system of the air compressor to make the cooling system of the air compressor in different overheat ranges, and continuously collect the operation state parameters of each valve under the current working state of the air compressor; Step 1.6: Replace some worn parts, and collect the operation state parameters of each valve on the air compressor under different fault types for different fault phenomena of the air compressor; Step 1.7: According to the operation state parameters of the valves collected in Step 1.1 to Step 1.6 and the corresponding operation state of the air compressor, establish sample data including the working state of the air compressor, fault types and the corresponding operation state parameters of the valves, and form a sample set.

2. The method for diagnosing multi-valve faults of an air compressor based on deep learning according to claim 1, wherein, The specific description of Step 2 is as follows: Step 2.1: Input the operation state parameters of each valve on the air compressor when a fault occurs, and realize the learning process of the deep learning model according to the operation state parameters under the valve fault; Step 2.2: Select the operation state parameters of a single valve fault on the air compressor and input them into the deep learning model. Replace the faulty valve in turn to ensure that the operation state parameters when all valves fail are input into the deep learning model, and the deep learning model can complete the recognition, and perform the single-layer training process of the deep learning model; Step 2.3: Select the operation state parameters of any number and any position of valves on the air compressor when a fault occurs and input them into the deep learning model. Select multiple groups of different valves and different operation state parameters and input them into the deep learning model, and the deep learning model can complete the recognition, and perform the multi-layer training process of the deep learning model; Step 2.4: Continue the training, collect the operating state parameters of the valves during the actual operation of the air compressor, adjust the working state of the air compressor valves, use the deep learning model trained in Step 2.3 to identify and collect the current operating state parameters and their corresponding fault types, and perform the reinforcement training process of deep learning. After the reinforcement training, a fault detection model is obtained.

3. A fault diagnosis method for multiple valves of an air compressor based on deep learning according to any one of claims 1 to 2, characterized in that, For each air compressor, collect the operating state parameters of each valve on the air compressor and the corresponding operating state of the air compressor, and construct a sample set for training the deep learning model as the fault monitoring model of the air compressor.

4. A multi-valve fault diagnosis system for air compressors based on deep learning, which is implemented based on the multi-valve fault diagnosis method for air compressors based on deep learning described in claim 1. A plurality of valves are installed on each air compressor, and it is characterized in that, The system includes a computer, a control center. The valve is electrically connected to the computer, and the computer is wirelessly connected to the control center; The computer is used to collect the operating state parameters of the valves on each air compressor and transmit them to the control center; The control center is used to construct a fault monitoring model based on the deep learning model.

Images