Fault Diagnosis Of Marine Main Engine Cylinder Cover Based On Support Vector Machines

During the development of fault diagnosis, to improve the rate for correct diagnosis has always been the hot subject. For elaborating systematically in mechanical fault diagnosis, the definitions and classifications were presented in the paper. Based on the inference process of fault diagnosis, the mechanism and possible causes were analyzed in detail.The cylinder pressure was determined from the vibration signal of a diesel engine cylinder cover by means of the wavelet analysis and time-series methods. The heat release rate was calculated by analyzing the indicated pressure diagram. This paper combines the cycles simulating calculation of the engine working process with the research on heat release rate, thus accomplishing the research on predicting the performance of a diesel engine under variable operating conditions according to the vibration signal of the diesel engine cylinder cover.Fault diagnosis is essentially a kind of pattern recognition, or classification. SVM is a valuable pattern-recognition method in theory and in application. Since the resolution principle of SVM is a quadratic programming whose local optimal value is just its global value, it averts falling into local optimal value in training and recognition. It is based on structural risk minimization principle and can balance optimally the complexity of algorithm (VC dimension) against the training accuracy (structural risk minimization), so it has good generalization ability.Support vector machine (SVM) is a novel machine learning method based on the statistical learning theory and the structural risk minimization principle. It is widely used in the field characterized by small sample, nonlinearity, and local minima, and has high generalization. An overview of the basic ideas underlying SVM for regression and function estimation is presented in this paper. The application of SVM to find the fault diagnosis of marine diesel engine cylinder cover is discussed. A simulation example is taken to demonstrate the effectiveness of the proposed approach.In this paper, the author discusses the applications of SVM in the fault diagnosis of marine main engine cylinder cover, include 5 chapters. Chapter 1 is mainly introduce about the theory and basic principle of Support Vector Machine(SVM). It have become one of the most popular approaches to learning from examples and have many potential applications in science and engineering. It is a new general machine-learning tool based on structural risk minimisation principle that exhibits good generalisation even when fault samples are few. SVM has shown powerful ability in learning with limited samples. As a direct implementation of the structure risk minimization (SRM) inductive principle, SVM provides good performances such as global optimization and good generalization ability. It has been developed as robust tool for classification and regression in noisy, complex domains. The two key features of SVM are generalization theory, which leads to a principle way to choose a hypothesis; and, kernel functions, which introduce non-linearity in the hypothesis space without explicitly requiring a non-linear algorithm.Now is the most advanced machine learning algorithm in the field of the pattern recognition ,and its characters have already showed more superiority than other methods, it can solve small sample learning problems better by using SRM than Empirical Risk Minimization. Moreover, by using the Kernel function idea, this theory can change the problem in non-linearity space to that in the linearity space in order to reduce the algorithm complexity.Chapter 2 is introduced about the Detection of Troubles in Marine Diesel Engines by Sound and Vibration. The vibrations and sounds of a diesel engine are very complex because of the engine's reciprocating movement and the variations in its operating cycle. It is therefore very difficult to detect mechanical troubles by these signals.This study makes it clear that in many cases, the time domain data monitoring of high frequency signals of over 20 kHZ or of fairly high order natural frequencies of the system are effective detection methods. The study also shows that detection by vibrational signals is much superior to that by acoustic signals.These two kinds of relatively high frequency signals, which are not so large during normal operation, greatly increase during abnormal operation to a greater extent than the low frequency signals or the low order natural frequencies of the system. Therefore, it becomes somewhat easier to detect trouble by monitoring such high frequency signals.So, new or improved monitoring methods for the mechanical parts of diesel engines are required. This chapter discusses some methods of trouble detection by sound or vibration. Chapter 3 is explained about the Content of Marine Diesel Engine Cylinder Cover fault diagnosis research. The increasing demand of quality in production and application processing has encouraged the development of several studies on fault detection and diagnosis in industrial plant. Most manufacturing processes involve many correlated variables. When any one of these variables deviates beyond their specified limits, a fault may occur. A quick and correct fault diagnosis system helps to avoid product quality problems and facilitates preventive maintenance.The fault diagnosis system should perform two tasks, namely fault detection and fault diagnosis. The purpose of the former is to determine that a fault has occurred in the system. To achieve this goal, all the available information from the system should be collected and processed to detect any change from nominal behaviour of the process. For example: temperature, pressure, flow rate, vibration level, noise, etc. The second task is devoted to locate the fault category or the fault source. In rotating machinery, the root cause of faults is often faulty rolling element bearings. One way to increase operational reliability of machine is to monitor and diagnosis incipient faults in these bearings.When an engine component participating in the combustion process of an internal combustion piston engine, malfunctions, this malfunction may be reflected in the ensuing cylinder pressure traces, exhaust valve, acoustic emission and vibration signals.Chapter 4 is the main part of paper, the signal characteristic and the typical fault diagnosis for Diesel engines cylinder covers' vibration apply by SVM. The vibration signatures of marine main diesel engine contain valuable information on the health of the combustion chamber components. It could be used to detect incipient faults in the engine. Several commonly occurring faults were induced in a 4-stroke diesel engine and the ensuing vibration signals recorded. The working conditions of diesel engine components can be detected by monitoring the cylinder cover vibration signals.The characteristics of cylinder cover vibration signals excited by valve impacts and combustion forces are discussed, and some criteria suitable for diesel engine vibration monitoring and fault diagnosis are proposed. By applying SVM method, signal models and parity equations residuals are generated. Detectable deflections of these residuals lead to faults.According to the result, SVM has been successfully applied to many applications, such as pattern identification, multi-regression, non-linear model fitting, etc. The results give the evidence that the technique is not only quite satisfying from a theoretical point of view, but also can lead to high performance in practical applications.Chapter 5 Conclusion and Future Work. The new approach to the diagnosis of an analog filter, based on the application of SVM has been presented. The diagnosis is understood as the recognition of the individual fault of element, where the fault means the change of the value above or below the assumed tolerance of elements. The important feature of the proposed solution is high accuracy and great speed of operation. Once the network has been trained, the recognition of fault is achieved immediately, irrespective of the size of the circuit. Thus the solution is suited for real time applications for fault location. The distinct advantage of the SVM is its good generalization properties .Trained on only limited number of representative examples of each fault, the network is able to recognize the non-ideal fault in the wide range of changed parameters and at some tolerance of elements. The accuracy of fault recognition checked on the example of bi-quadratic filter by using SVM is better than those presented recently for similar tasks.The experiment results in this work indicated that the SVM method has better effectiveness than traditional artificial neural network and other methods. The multi-class faults classifier designed in this study has many advantages: simple algorithm, good classification and high efficiency. It is very suitable for online monitoring and diagnosis. SVM provides us a new and useful method for developing intelligent diagnosis.This method can improve the robustness and accuracy of fault diagnosis. Furthermore, they can also be applied in other field.Future research will focus on the selection of optimal sensor suite for maximizing diagnosibilty. We plan to perform Fault Detection and Diagnosis (FDD) based on extracted features. This will reduce computational complexity, and will also facilitate FDD using transient data. We also plan to develop a more robust some modeling to aid in future Marine Diesel Engine Fault Detection and Diagnosis.In future, we plan to conduct more experiments on SVM with multi-class documents and a large number of single class documents with different forms of feature representation.Finally, we can say that SVM have become the hot spot of machine learning because of their excellent learning performance. They also have successful applications in many fields, but as a new technique, SVM also have many shortcomings that need to be tracked and bettered ,including: the adaptive kernel and parameter selection, the shortcomings of training methods and incremental learning, etc. because of these problems, the applications of SVM are limited in many fields.Analysis of the experimental results shows that four kernel functions have the same prediction performance in short term prediction and radial basis function (RBF) kernel has better prediction performance than other kernels in long term prediction. The normalized mean square error (NMSE) of RBF kernel decreases by about 20% compared with other kernels. In comparison with conventional back propagation neural network (BP), radial basis function network(RBF) and generalized regression neural network(GRNN), the results show that SVM, which implement the SRM principle, obtain the best prediction performance and the NMSE of SVM decreases by about 15% in long term prediction.In this paper, the author discusses the applications of SVM in the fault diagnosis of marine main engine cylinder cover. The author carries on simulation research and experiment research of fault diagnosis by using SVM theory, including design of the mathematics model, simulation model, extraction of the feature parameter and data collection of the experimental system.