機器運轉時都會伴隨著振動,而當機器振動過大時,則表示可能發生故障現象,對於馬達而言,在其發生故障時,可透過多種振動分析方法加以診斷,而每一種分析方式,皆有一定鑑別的特性,若使用單一分析方法,有時未能正確診斷出故障原因,為了能提高故障鑑別率,本文提出以頻譜、頻瀑及軸心軌跡圖分析為主之正反向綜合推理診斷方法。在正向推理中頻譜及頻瀑分析,應用了倒傳遞類神經網路的理論及子網路的概念,分別架構出轉子、軸承、電機三個子網路,透過類神經網路診斷的方式做正向推理,求取故障可能發生原因。軸心軌跡正向推理方式,利用人為識別軸心軌跡圖形,以圖形判別可能的故障類型。反向及綜合推理方式,是利用貼近度與正向推理結果進行加權支持度計算,其中貼近度應用海明(Hamming)距離之貼近度理論,將提取之故障特徵信號與定義之實際故障時所對應故障特徵量做貼近度計算。 依據正反向綜合推理理論,透過三個實際例子進行驗證,從正向推理、反向推理及綜合推理方式的輸出診斷結果證明正反向綜合診斷方法的合理性、可靠性與準確性。
The machine runs with vibration. If the machine vibrates oversized, it maybe has the fault. Regarding the motor fault diagnosis, we can use many kinds of vibrations analysis method to diagnose. Each vibrations analysis method has different characteristics. Sometimes using the sole analysis method, we can't correctly diagnose the fault. In order to raise the rate of diagnosis ability, this article proposes the pro and con inference and the mix inference method which include frequency spectrum analysis, waterfall analysis, and orbital analysis. In positive reasoning, the frequency and waterfall analysis apply the Back-Propagation Neural Network (BPNN) and subnet theory. We establish separately the rotor, the bearing, and the electrical machinery neural network. We use the neural network to inference the possible fault from the reason. Orbital analysis means to use artificial recognition orbit by graph to distinct the possible fault type. The negative inference and the mix inference ways use the approximate reasoning and positive reasoning result with weighting computation to the inference result. The approximate reasoning method uses Hamming distance theory. We calculate relevance between experiment signal data and define fault signal data by the approximate reasoning method. Based on the pro and con inference and the mix inference method, we develop the professional system of motor fault diagnosis. We also test for three practical examples to do positive reasoning, negative reasoning and mix reasoning. From each inference way, output data proves that mix inference diagnosis method is reasonable, reliable and accurate.