Gear Box Research Paper

ON-ROAD TESTING OF A VEHICLE FOR GEARBOX FAULT DETECTION USING VIBRATION SIGNALS

ABSTRACT
Gearbox is one of the most important components in an automobile, enabling power transmission from the engine to the wheels. Gears and bearings are prone to failure. The impending case of failure can be predicted by performing vibration analysis of a gear box, usually done by acquiring data in lab conditions. This paper deals with the condition monitoring of a gearbox in real time test conditions. The information from the data is extracted in the form of statistical features from the normalized raw data. The classification ability of the decision tree algorithm is discussed to automate the gearbox fault diagnosis.

Keywords: Gearbox, Real time testing,

Decision tree is a commonly used classifier. The decision tree algorithm is simple, non-parametric, computationally faster and easy to interpret. Decision tree classification has been done in vibration based gearbox fault diagnosis and was found to give good results. Decision tree has a great potential in deriving the rules from the feature set [18]. Decision tree reduces the domain knowledge required by identifying the relevant features [19]. Decision tree consists of number of branches starting from a root and ending in number of leaves. This tree like structure is induced with nodes wherein each node acts as an input feature. Here the root represents the feature with the highest weight and the leaf represents the irrelevant features or the features with the least weightage. Figure-8 shows a part of the decision tree formed by using raw data. In this decision tree kurtosis has the highest weightage and the weightage decreases as we go down the tree. Similarly Figure-9 shows a part of the decision tree formed using normalized data. Variance was found to have the maximum weightage in this case. Thus decision tree shows a descending order of weightage from the root to the leaf. The features extracted from the time-domain data were used to train a decision tree and the classification efficiency was

A part of the Decision tree using normalized data.
RESULTS AND DISCUSSION
Test has been carried for the conditions as mentioned in Table 1 and the time-domain data acquired were stored. The time-domain data acquired from the accelerometer for different test conditions were then normalised. This huge data acquired for each test condition was then extracted with the help of variables such as mean, median, mode, variance, skewness, crest factor, root mean square (RMS), kurtosis, RMS*kurtosis, peak value, standard error and was stored.
The extracted data corresponding to different conditions of the gearbox for 1st gear tested at 10kmph were combined to a single file. A decision tree was trained using the combined data and the classification efficiency was determined. Table 3 and Table 4 represent the test results in the form of a confusion matrix when the data was processed without and with normalisation respectively. In Table 3 and Table 4, GG represents Good Gearbox, FG represents Faulty Gear, FB represents Faulty Bearing, N represents Normalised and WN represents Without Normalisation

Figure-10. Data Representation.

Table-5. Confusion matrix for Gear 1 tested at 10 kmph (without