Mechanical Fault Prediction Model And Its Application Based On Grey Theory And Neural Network

With the improving performance and the increasing complexity of heavymechanical equipment systems, fault prediction and fault diagnosis are more and morevalued by people. Because the traditional fault diagnosis technology has been unable tomeet the need of the actual operation of modern complex mechanical equipment,resulting in serious or catastrophic accidents still occur frequently for mechanicalequipment malfunction. How to real-time condition monitoring and fault prediction formechanical equipment system, in order to achieve maintenance equipment known inadvance, and ensure that equipment is zero fault operation, has become the focus offault prediction research. And the key technology to achieve the goal is mechanical faultprediction technology. Therefore, study on the fault prediction techniques of mechanicalequipment systems, whether it is to reduce economic losses, or extend equipment life,has the vital significance.The prediction model is one of the core contents of the mechanical fault predictiontechnology. As a result of mechanical equipment system usually have the characteristicsof nonlinear, uncertainty, dynamic time-variability and Small sample data, how toconstruct the prediction model satisfying the actual running of the equipment system isneeded to solve the primary problem in fault prediction technology. This paper willfocus on this problem, taking the project of Chongqing scientific technology "heavyequipment intelligent preventative maintenance and automatic control system" as thebackground, the fault prediction model is studied further. The main research content andconclusion include:(1) Through studying further on GM(1,1) prediction model, from the view ofimproving smooth degree of discrete data sequence, this paper proposes a newimproved GM(1,1) model based on function cot(x~A)(A>1)transformation.According to the gray modeling theory, this paper first proposes the sufficientconditions that transformation of function cot(x~A)can improve smooth degree ofdiscrete data sequence more effective than cot(x)function method and cot(xα)function method. It has been proved theoretically that this function transformationsatisfying the proposed sufficient conditions can improve effectively smooth degree ofdiscrete data sequence, and has better smoothness than other existing transformationfunctions. On this basis, the conclusion is generalized to more general cases, and this paper further gives out an optimized method of determining parameter A to improvesmooth degree of modeling data sequence more effectively and applies this functiontransformation in GM(1,1) model. So it greatly extends the application scope ofGM(1,1). The experimental results show that not only can this function transformationimprove the smooth degree of modeling data sequence, but also improve simulationaccuracy of the proposed model. Therefore, the effectiveness and practicality of theproposed method is verified.(2) As to the characteristics of fault behavior for mechanical equipment systemwith the nonlinear, uncertainty, dynamic time-varying and small samples of collecteddata, a dynamic failure prediction model based on multi-parameter grey error term andneural network is established. First, the small sample of the collected data to conduct acumulative, MGM(1,n) prediction model is developed using the monotonicallyincreasing sequence to obtain the initial prediction values of the original sequence. Thenthe modified values of the initial prediction values is predicted using BP neural network,the final prediction value is equal to the initial prediction values plus its modified value.Simulation results show that the proposed prediction model has a high predictionaccuracy and broad application prospects.