TWI403748B - Motor Fault Diagnosis Method and Device Based on RBF Information Fusion Technology - Google Patents

Abstract

The invention provides a diagnosis method and a device for dysfunctional motor using RBF information integration technique. The method includes setting up various sensors on the motor structure, obtaining functional relationships among the sensors through RBF neural network, predicting output status of any one of the sensors, obtaining a RBF training algorithm, establishing an integrated testing model based on prediction signals and measured signals of the sensors, and diagnosing motor dysfunction through voting integrated testing policy. The device includes a sensor, a network module, a RBF calculation module and information integration calculation module. The invention can perform information integration calculation through RBF neural network to increase the speed and accuracy of diagnosis of dysfunctional motors.

Description

Motor fault diagnosis method and device using RBF information fusion technology

The invention relates to a motor fault diagnosis method and device, in particular to an application RBF information fusion technology, which has a hardware architecture for multi-sensor information fusion calculation, and uses RBF multi-sensor information fusion calculation to accelerate calculation motor fault. An innovative designer of diagnostic performance and correctness.

In recent years, due to the rapid development of industry and the increasingly complex system architecture, it is relatively difficult for maintenance personnel to find system faults. Therefore, in order to save time for inspection and increase the accuracy of detecting fault points, review and Improve the quality of production, thus opening up the research field of fault diagnosis.

The motor has been around for more than 100 years. It is one of the mothers of industrial power and one of the main sources of power. Its structure and function are quite mature. At present, the pace of research and development by experts and scholars has shown that it is difficult to make major breakthroughs in the short term, however, The user often faces the problem of motor failure. Of course, the failure of the motor is caused by improper use or damage to the structure. However, when the motor cannot be grasped and monitored in advance, when the motor fails, Only after waiting for the motor to be updated or repaired can the problem be solved, and the production schedule is stopped and the working hours are delayed, which causes the user to suffer heavy losses.

In the past, the research on motor fault diagnosis mostly focused on the diagnosis and analysis of abnormal vibration in each detail position, ignoring the key cause of the failure of another motor, that is, the temperature rise phenomenon of the motor; the motor temperature is too high, the heat dissipation effect is not good, the current If it is too large, exceeds the load, etc., when the above reasons occur, the degree of affecting the motor is more than the vibration, so it is obviously necessary to be included in one of the motor fault diagnosis items.

In view of this, the inventor has been engaged in the manufacturing development and design experience of related products for many years. After detailed design and careful evaluation, the inventor has finally obtained the practical invention.

The main purpose of the present invention is to provide a motor fault diagnosis method and device using the RBF information fusion technology, and the problem to be solved is to develop a new motor fault diagnosis method and device which has ideal utility. Thinking about the breakthrough of the target; the technical feature of the invention solves the problem, in terms of the method aspect, mainly sets up multiple types of sensors on the motor structure; and uses the radial basis function-like neural network (RBFNN) to obtain each The functional relationship between the sensors is used to predict the output state of each sensor, and a training algorithm of radial basis function is obtained; then the fusion detection is established by using the predicted signals of the sensors and the measured signals as input credentials. The model adopts the voting fusion detection criterion to complete the motor fault diagnosis; and in terms of the structural aspect, the motor fault diagnosis device comprises the sensor, the network module, the radial basis function calculation module and the information fusion calculation The module is constructed; by this design, the invention mainly uses the radial basis function-like neural network architecture to perform information fusion calculation, thereby achieving To the practical advancement of accelerating the immediacy and correctness of motor fault diagnosis.

Please refer to the first, second, and third figures, which are preferred embodiments of the motor fault diagnosis method and apparatus using the RBF information fusion technology of the present invention, but the embodiments are for illustrative purposes only, and are not used in patent applications. Limited by this structure; the motor fault diagnosis method includes:

a. arranging a plurality of types of sensors on the motor structure; the sensors may include temperature sensors, vibration sensors.

b. Using a radial basis function-like neural network (RBFNN) to obtain a functional relationship between the sensors on the motor, thereby predicting the output state of each of the sensors, and obtaining a radial basis function training Algorithm.

c. Using the predicted signal and the measured signal of each sensor as input credentials, a fusion detection model is established to perform data fusion calculation, and the voting fusion detection criterion is used to complete the motor fault diagnosis.

The motor fault diagnosis item may include items such as fault sensor addressing, fault type identification, fault degree judgment, and fault sensor normal output estimation.

Next, the motor fault diagnosis device using the RBF information fusion technology disclosed in the present invention, as shown in FIG. 2, includes the following components: a complex sensor 10 for sensing different operating states of the motor 05 structure. The sensor may include a temperature sensor 11 and a vibration sensor 12.

A network module 20 uses a network interface to transmit or receive information obtained by each of the sensors 10; a radial basis function calculation module 30 is implemented on an embedded integrated development environment (NIOS IDE) Radial Basis Function Neural Network (RBFNN) algorithm, and convert it into hardware acceleration calculation using C2H compiler; an information fusion computing module 40 is used to process information hierarchy, information fusion method process and related Data; thereby, the motor fault diagnosis device can implement radial basis function calculation on the embedded processing platform, and utilize the SOPC programmable system chip method to design information fusion into the processed information level and information fusion method.

The main core design of the invention is to apply the wireless sensing network to detect the vibration and temperature rise of the motor. Since the wireless sensing network is easy to set up, several sensors 10 can be set up in different parts of the motor, and the sensors are respectively 10 The signal of vibration and temperature rise is transmitted back to the computer, and the computer performs fast Fourier operation and converts into a frequency domain signal, and compares the frequency domain signal with the self-established fault type, and after being classified by the membership function, The radial basis function-like neural network inference completes the diagnosis. Finally, the diagnosis results are clearly displayed on the computer for the administrator to view. The neural network has the special ability of automatic learning, experience accumulation, inference association and inductive judgment. It can effectively improve the speed and accuracy of motor fault diagnosis, and achieve the function and practical progress of mastering and monitoring the instantaneous condition of the motor.

Among them, the Radial Basis Function Neural Network (RBFNN), also known as the radial base function neural network, is mainly characterized by the local adjustment function of simulating the axons of the cerebral cortex. Very good mapping capability, its architecture is similar to Multilayer Perceptron (MLP), with one input layer, one hidden layer and one output layer. It is a multi-layer feedforward neural network in the system architecture of neural network. MultilayerFeedforward Networks has the advantage of greatly reducing the learning process time. The radial basis function neural network generally adopts a two-stage learning rule in the learning process, and the pre-stage training uses unsupervised learning ( Supervised Learning), the post-stage training is Unsupervised Learning. The so-called supervised learning is to give a neural network training paradigm in the learning process. Each training paradigm contains both input and target output. The value (as shown in the figure above in Figure 4), this target output value plays the role of the teacher Color, constantly urging the neural network to correct the weight value throughout the training process, by re-adjusting the strength of the network weight value again and again, to reduce the gap between the actual output value of the network and the given target output value, Training will not stop until the gap is less than a certain threshold or the weight value is no longer changed. Non-supervised learning only needs to provide input information, and does not need to expect output information, that is, it does not need the gap between the actual output value of the network and the target output value to improve the output value of the neural network, that is, It is said that the weight value can be learned and adjusted according to the characteristics of the input information (as shown in the figure below in Figure 4).

The so-called neural network, the structure of which is shown in Figure 5, has three layers, from left to right, respectively, the input layer, the hidden layer and the output layer. The multi-layer feedforward neural network in the feed-like neural network (Feedforward), the concept of the radial basis function-like neural network is to construct a large number of radial basis functions, found by the function approach method (Curve Fitting) The mapping between input and output. In general, the radial basis function of a hidden layer neuron in a radial basis function-like neural network can be divided into seven types. The present invention uses a Gaussian function as a hidden layer neuron. The function, with the input values of N dimensions, the hidden layer of M neurons, and an output value as representations of the architecture and algorithm of the radial basis function-like neural network.

Among them, the means of "information fusion" is to intelligently synthesize multi-source information from a certain target to generate more accurate and complete estimation and judgment than a single source through multi-sensor information fusion, which can expand the space-time coverage. Range, increase confidence, improve the reliability of the detection system; according to the information level information fusion method handled by the system, it is generally divided into three types: information layer fusion, feature layer fusion and decision layer fusion. There are many specific implementation technologies for information fusion. This author uses neural networks to study the two methods of feature layer fusion and decision layer fusion.

The invention applies the neural network to the data fusion application of the wireless sensing network, so that the system can flexibly, accurately, reduce redundant information and low-cost wireless data acquisition and transmission, and improve the anti-interference of the data acquisition system. Ability and efficiency. Moreover, the present invention provides a neural network integration algorithm for adaptive learning in the fusion calculation of data.

Advantages of the invention:

The invention mainly utilizes the latest radial basis function neural network for information fusion calculation to accelerate the immediacy and correctness of motor fault diagnosis. In recent years, wireless sensing networks have flourished. Information fusion computing is widely used in the design of multi-sensor network environments. It is very slow to use traditional information processing methods to diagnose motor faults. I hope that in wireless sensing networks. The calculation of radial basis function neural network on the platform and application to motor fault diagnosis will improve the efficiency and correctness of motor fault diagnosis.

The invention can improve the calculation time of current motor fault diagnosis, complete the fault position and property accuracy early, shorten the fault judgment time, and enhance the receiving signal capability of the wireless multi-sensor environment, and can be applied to the motor electronic related industry. And has excellent industrial utilization.

The above embodiments are intended to be illustrative of the present invention, and are not to be construed as limiting the scope of the invention. The principles are changed and modified to achieve an equivalent purpose, and such changes and modifications are to be included in the scope defined by the scope of the patent application as described later.

05. . . motor

10. . . Sensor

11. . . Temperature sensor

12. . . Vibration sensor

20. . . Network module

30. . . Radial basis function calculation module

40. . . Information fusion computing module

Fig. 1 is a block diagram showing the method of motor fault diagnosis according to the present invention.

Fig. 2 is a schematic view showing the structure of the motor fault diagnosis apparatus of the present invention.

Fig. 3 is a block diagram showing the operation logic of the motor fault diagnosis method of the present invention.

Fig. 4 is a block diagram of the text of the radial basis function neural network learning rule of the present invention.

Figure 5: A simplified diagram of a neural network architecture such as the present invention.

05. . . motor

10. . . Sensor

11. . . Temperature sensor

12. . . Vibration sensor

20. . . Network module

30. . . Radial basis function calculation module

40. . . Information fusion computing module

Claims (3)

A motor fault diagnosis method using the RBF information fusion technology includes: a. setting a plurality of types of sensors on a motor structure; wherein the sensors include a temperature sensor and a vibration sensor; b. utilizing a radial direction A function-based neural network (RBFNN) acquires a functional relationship between the sensors on the motor to predict the output state of each of the sensors and derive a training algorithm for the radial basis function; c. Taking the predicted signal and the measured signal of each sensor as input credentials, a fusion detection model is established, and the voting fusion detection criterion is used to complete the motor fault diagnosis; wherein the motor fault diagnosis includes fault sensor addressing and fault. Type identification, fault level decision, and fault sensor normal output estimation. A motor fault diagnosis device using RBF information fusion technology includes: a plurality of sensors for sensing different operating states of the motor structure; and a network module for transmitting or receiving each of the sensors by using a network interface Information; a radial basis function calculation module, implemented in a embedded integrated development environment (NIOS IDE) on the radial basis function (RBF)-like neural network algorithm, and converted to hardware using the C2H compiler Accelerated computing; an information fusion computing module for processing information levels, information fusion method processes and related data; Thereby, the motor fault diagnosis device can implement radial basis function calculation on the embedded processing platform, and utilize the SOPC programmable system chip method to design information fusion into the processed information level and information fusion method. A motor fault diagnostic apparatus using the RBF information fusion technology according to claim 2, wherein the sensor comprises a temperature sensor and a vibration sensor.