Built-in intelligent fault diagnosing device based on data inosculating pattern recognition and method thereof

[0031] The intelligent diagnostic device for rail transit train embedded faults based on data fusion pattern recognition includes hardware and software. Both hardware and software adopt a functional modular design. The hardware structure is shown in Figure 1, including a data acquisition module and an embedded module. It consists of a signal acquisition unit and a central processing unit. The signal acquisition unit includes a sensor, signal conditioning, AD conversion, DSP (or MCU), digital input module, FLASH memory and CAN bus interface. The signal acquisition unit is directly installed and fixed on the main parts of the train's key equipment to complete the fault diagnosis, calibration and calibration of the sensor, train equipment working state parameter acquisition, signal feature extraction and CAN bus communication. The central processing unit uses an embedded PC with the PC104 bus. PC104 is an industrial control bus specifically defined for embedded control. It is essentially a compact IEEE-P996 standard, and its signal definition is basically the same as PC/AT. But the electrical and mechanical specifications are completely different. It is an optimized, small, stacked embedded system with excellent vibration resistance. The PC104 embedded computer module series is a complete set of low-cost, high-reliability, and structured modules that can be quickly configured into products. Meet the requirements of system vibration resistance and reliability. The central processing unit is located in the monitoring center of the equipment, and mainly completes the processing of the collected signals, the judgment and alarm of the working status, and the fault diagnosis.

[0032]The fault diagnosis device uses a high-speed DSP microprocessor, with the help of its high-speed data processing multi-level buffer and pipeline mechanism, to form a distributed monitoring and processing system, that is, the upper and lower computer two-level microcomputer structure. Among them, the data acquisition module is responsible for completing the tasks of data acquisition and signal processing, including the collection of sensor output signals and the preprocessing of sensor data; while the embedded module realizes the functions of data storage, graphic display and remote data transmission. In addition, the advanced embedded system structure is adopted to realize the functions of data storage, graphic display and data remote transmission, and complete the task of fault diagnosis. The use of this embedded fault intelligent diagnosis technology can meet the requirements of the system's speed, volume, power consumption, etc., thereby improving the safety and reliability of the equipment, and maximizing the efficiency of the equipment.

[0033] Figure 2 is the software structure of the invention. The software system mainly includes a monitoring database, a monitoring alarm module, a fault diagnosis module, a CAN communication module and a network communication module.

[0034] Figure 3 is a flowchart of the software module program. The program flow of the software module is to first carry out CAN, network initialization, preprocess the read data, transfer it into the network knowledge base for reasoning, and can locate the fault? If the fault can be located, display the result, otherwise enable remote diagnosis? If yes, perform remote diagnosis, if not, list the cause of the failure according to the probability, and display the result to store the failure parameters. Monitor and alarm the diagnostic fault parameters. The monitoring and alarm module adopts single parameter threshold alarm and multi-parameter comprehensive analysis alarm to forecast the current system status according to the current sensor signal. Single parameter alarm compares the monitoring data of a single working condition parameter with its set normal working condition threshold, and performs hierarchical alarms according to the degree of difference; multi-parameter alarm first normalizes the monitored values ​​of several key working condition parameters , And then carry out comprehensive fault diagnosis, and give system-level status indication. The monitoring alarm module divides the alarm into 3 levels according to the severity of the specific working conditions. Level 1 alarm: It is just a reminder that the equipment can still work for a period of time in this case; Level 2 alarm: It is required to pay close attention to the development of the fault, but does not need to be dealt with immediately; Level 3 alarm: For this level of alarm, an emergency should be carried out immediately deal with.

[0035] Fault diagnosis module, the complexity of the equipment structure makes its faults have the characteristics of multi-level, fuzzy and uncertainty, and this system requires online diagnosis, so the fault diagnosis module adopts the neural network diagnosis model. When the fault diagnosis module works, the data collected by the lower computer of the monitoring database is taken out and preprocessed as the input of the neural network, and the output of the neural network represents the result of the fault diagnosis. First, the neural network is trained and learned, that is, the state parameters corresponding to the specific fault are used as samples to establish a relatively complete sample library, and then use all the samples to train the neural network, so that the knowledge of the sample library can be stored in the form of a network In the connection right of the neural network, finally, the fault diagnosis can be completed through the calculation of the input of the neural network. Most of the models used by neural networks in fault diagnosis are BP models. This is mainly because the research on BP models is relatively mature and reliable in use.

[0036] Figure 4 is a diagnosis system based on the data fusion method for the fault diagnosis module. The system includes three modules: a data-level fusion module, a feature-level parallel multi-neural network local diagnosis module, and a decision-level fusion diagnosis module; the data-level fusion module mainly performs multi-sensor data collection and feature extraction. In order to perform fault diagnosis, the main parameters needed are monitored and obtained from the test bench through multiple sensors, and converted into digital signals through D/A and A/D conversion and input to the computer for system monitoring and diagnosis. At the feature level, three parallel BP (back propagation) neural networks with the same structure are used. The BP neural network is a relatively mature neural network at present. The present invention adopts the most basic three-layer BP algorithm; the decision-making level merges the local diagnosis results of the BP network to obtain the final diagnosis result. The flow of the diagnostic program is shown in Figure 5. Proceed as follows:

[0037] 1) Data collection and storage in the database;

[0038] 2) Read characteristic data from the database;

[0039] 3) Judge whether the feature extraction is completed, if it is not completed, go back to step 2; if it is completed, go to step 4;

[0040] 4) Send the read characteristic data to the neural network, and conduct multiple neural networks for training, reasoning, and fault location;

[0041] 5) Read the trained network connection weight, and calculate the network output;

[0042] 6) Perform fusion calculation on any two neural network outputs;

[0043] 7) Integrate the fusion result with the third network;

[0044] 8) Judge the result of the fusion diagnosis in step 7, whether to continue the diagnosis, if yes, return to step 1, repeat steps 1 to 7, otherwise end.