【摘要】：模擬電路故障診斷研究已有數十年歷史,受元件容差、非線性、溫漂等因素影響,該課題一直是研究的難點和熱點。電子器件中,模擬電路所占比例不大,但故障問題出現最多,電子器件運行可靠性很大程度上依賴于模擬電路的可靠性。傳統方法在電路規模日趨增大的背景下表現出數據處理能力弱、診斷時間長、診斷過程復雜等局限性,人工智能方法特別是神經網絡方法為其提供了新的研究方向,能很好適應非線性電路診斷,不依賴具體電路,降低診斷難度,但在診斷時間、診斷精度、糾錯容錯方面仍表現不足,并且建模過程復雜。基于覆蓋理論構造性神經網絡是近年來提出的新型神經網絡方法,它相比于傳統神經網絡具有建模簡單、魯棒性好、運算能力強的優點,適用于海量數據、復雜環境等情況下的工業應用,特別能大大降低運算時間。本文以神經網絡理論為基礎,克服現有故障診斷系統需要提取故障特征,故障建模過程復雜,系統運行中難以實現知識擴充等問題,提出將構造性神經網絡方法應用于模擬電路故障診斷中,取得良好的診斷結果。本文首先以M P神經元球面模型為基礎,建立基于球面領域的構造性神經網絡,對模擬電路具有±4%擾動故障樣本進行診斷能達到100%診斷精度;然后針對具有±15%擾動樣本某些故障無法診斷問題,通過設定拒識模式并通過增加神經元方法對無法診斷故障進行學習擴充,重新訓練神經網絡,能對新故障完全診斷并提升整體診斷精度;針對實際工業應用中需要處理海量數據,診斷系統存在優化約簡的問題,本文采用領域覆蓋和模糊覆蓋算法對神經網絡進行優化構造,診斷范圍從最大軟故障擴大為所有軟故障模式,診斷精度分別能達到89.3%和94.9%,并且能降低神經元個數,減小計算難度、計算量,降低診斷時間,同時使用模糊覆蓋算法對最大軟故障模式進行診斷,單選診斷率為85.71%,三選能實現100%診斷。實驗證明本文方法具有很強容錯能力,泛化能力好,特別適合復雜環境下電路故障診斷,具有良好發展前景。
[Abstract]:The research on fault diagnosis of analog circuits has been studied for decades. Affected by element tolerance, nonlinear, temperature drift and other factors, this subject has always been a difficult and hot research topic. Among electronic devices, analog circuits account for a small proportion, but the fault problems occur the most. The reliability of electronic devices depends on the reliability of analog circuits to a large extent. Under the background of the increasing size of the circuit, the traditional method shows the limitations of weak data processing ability, long diagnosis time and complex diagnosis process. Artificial intelligence method, especially neural network method, provides a new research direction for it. It can adapt to nonlinear circuit diagnosis, does not rely on specific circuits, and reduces the difficulty of diagnosis, but it is still insufficient in diagnosis time, diagnosis accuracy, error correction and fault tolerance, and the modeling process is complex. The constructive neural network based on coverage theory is a new neural network method proposed in recent years. Compared with the traditional neural network, it has the advantages of simple modeling, good robustness and strong computing ability, and is suitable for massive data. The industrial application in complex environment and so on can greatly reduce the operation time. In this paper, based on the theory of neural network, the existing fault diagnosis systems need to extract fault features, the process of fault modeling is complex, and it is difficult to expand the knowledge in the operation of the system. In this paper, the constructive neural network method is applied to analog circuit fault diagnosis, and good diagnosis results are obtained. In this paper, based on the spherical model of M 鈮，

本文編號：2480995