METHOD FOR LOW COST ROBUST MULTIPLE FAULTS DETECTION IN AUTOMOBILE ENGINES BASED ON NEURO-COMPUTING APPROACH

摘要

ABSTRACT The challenging problem of multiple faults detection in automobile two stroke, four stroke and 800 cc car petrol engines is solved efficiently by using a simple method. This method employs a single sensor and neuro-computing approach, which constitutes signal processing, neural networks and Support Vector Machines. From sound recording of healthy and faulty engines, significant features are extracted using signal processing resulting into a knowledge base, which is partitioned into training, CV and Testing dataset. The efficacy of the estimated neural network models are rigorously evaluated, so that the proposed technique is biased neither for specific data nor for any partitioning scheme for obtaining reasonable classification accuracy for combined multiple faults. The reliability and consistency in faults detection by the adopted technique is fully justified as the recorded signal always has inherent embedded background noise. The invention is described by way of real world example with reference to the following drawings: Figure 1: Shows the Working of Proposed Fault Detection System, in which following are the sequence of operations. 1 denotes three different types of engines are considered for Fault Detection System, out of which one Automobile Engine is selected, 2 denotes the engine status normal and different faulty conditions, 3 denotes the recording of Healthy & all Faulty Signals from the Engine, 4 denote the completion of recording, 5 denotes the normalisation and conversion into discrete samples, 6 denotes the change in gear position and speed of the engine, 7 denotes the Knowledge base Representative of Healthy & Faulty Conditions of signals, 8 denotes the Design & Development of Statistical & ANN Based Classifiers, 9 denotes the Selection of Optimal Classifier, 10 denotes the checking of the optimal classifiers, 11 denotes end, 12 denotes no, 13 denotes yes, 14 denotes no, 15 denotes yes. Figure 2: Shows the working of statistical classifiers in which, 1 denotes the knowledge base of different faults is applied to Statistical Classifiers, 2 denotes the selection of classification and regression trees for single and multiple fault conditions, 3 denotes the selection of Qualitative Output and Quantitative Inputs from an Engine’ Data Comprising of Normal and Faulty Condition, 4 denotes the selection of Measure and Tree Depth for each method, 5 denotes the performance of the classifier, 6 denotes the confusion matrix of each classifier, 7 denotes the status of the classifier, 8 denotes the comparison of classification Accuracy of all classifiers, 9 denotes selection of the optimal classifier on the basis of best performance, 10 denotes the status of selection of optimal classifies, 11 denotes the end, 12 denotes no, 13 denotes yes, 14 denotes yes, 15 denotes no. Figure 3: Shows the Working of ANN Based Classifiers in which, 1 denotes the selection of Knowledge Data Base for ANN based classifiers for Single/Multiple faults, 2 denotes the data partitioning, 3 denotes the selection of classifiers amongst ten types of classifiers. (MLP, GFF, MNN, JEN, PCA, SOFM, RBF, TLRN, RN & SVM) to classify the faults, 4 denotes the application of training datasets to ANN based classifier, 5 denotes the testing data is applied to trained ANN based classifier, 6 denotes Training of an ANN based classifier, 7 denotes the checking of training process, 8 denotes the Freezing of Connection Weights & Biases for the Trained ANN based classifier, 9 denotes the examination of Classification Accuracy for Estimated Sample, 10 denotes the status of classifiers, 11 denotes the selection of the next classifier, 12 denotes the design and development of optimal classifier, 13 denotes the evaluation of optimal classifiers, 14 denotes the Validation of ANN based classifiers, 15 denotes the end of process, 16 denotes no, 17 denotes yes, 18 denotes no, 19 denotes yes. Figure 4: Shows the proposed architecture of Fault Detection System for two stroke engine, which incorporates an automobile engine under test which is surrounded by different faults i.e. Air Filter Fault (FF), Spark Plug Fault (SP), Rich Mixture Fault (RM), Gudgeon Pin Fault (GP), Insufficient Lubricants Fault (ISL) and Piston Ring Fault (PR) and + sign indicates extension of the system for any other additional faults. The other parts of architecture are: 1. Microphone as a sensor 2. Data acquisition system 3. Signal processing system 4. Decision Making Neural Network Committee and 5. Visual Display System Figure 5 and Figure 6: Also shows the proposed architecture of Fault Detection Systems for HHPFS and MSAFS engines respectively. The working of the architectures for HHPFS and MSAFS engines are similar to two stroke engine as shown in Figure.