Condition Based Maintenance Using Neural Networks

In this proposal, Intelligent Automation Inc. (IAI) proposes an innovative new method of engine degradation monitoring, fault detection, and diagnostics, which is robust to sensor noise and is efficient in training and learning. This innovation will also allow us to detect new fault conditions that have not occurred before. This may include sensor failures and hence, the capability of validity self-checks. We propose to use two neural networks: Principal Component Analysis (PCA) and Fuzzy CMAC. PCA is a powerful technique for extracting the features inside the input signals. We have seen that PCA can reduce the signal dimension from 800 to 5 in several applications. A major advantage of PCA is that supervised learning is unnecessary. It can also be implemented in low-cost hardware. The Fuzzy CMAC (Cerebellar Model Arithmetic Computer) inherits preferred features of arbitrary function approximation, fast learning, and parallel processing from the original CMAC neural network, and the capability of acquiring and incorporating human knowledge into a system and the capabilities of acquiring and incorporating human knowledge into a system and processing information from the fuzzy inference rules of fuzzy logic. Our learning rates are at least an order of magnitude faster than conventional neural nets. The Universal Approximation Theorem that we have formally proven shows that any function can be learned to any degree of accuracy with enough learning cycles. The proposed method is relevant to the subtopic because our innovation can provide an early warning before the system actually fails. The purpose is to increase time between overhaul and safety.