Diagnosis and detection of mechanical fault in an internal combustion engine at the end of line engine testing is very crucial to deliver customer a defect free healthy engine. Currently engine testing operator uses his subjective hearing expertise to diagnose and detect any abnormal noise coming out from the engine. Subjective engine testing leads to continuous noise exposure to operators, possibilities of defective engine go to customer due to hearing fatigue of the operator, there are few abnormal noises which are beyond human perception which sometime manifest itself when it is loaded on the vehicle and in few cases, there is bias among the operators’ perceptions as the testing is subjective in nature. More over operators needs rigorous training to become an expert of engine noise testing. The theme of the project is to convert the subjective engine testing to objective engine testing by developing an engine vibration measurement system and deploy a machine learning algorithm model which can detect the faulty engine and also diagnose the type of fault (Head noise, Gear damage, Whining noise etc.) in the engine with less or no manual intervention. The vibration measuring system consist of three unidirectional accelerometer sensors with magnetic mounting base which can be mounted on the three different locations of the engine. Vibration data is captured by a vibration analyzer hardware at three different engine speeds. Statistical method is used to extract the time domain and frequency domain features data and feature ranking method is used to reduce the dimensionality of the features. We have collected more than 550 engines vibration data combination of both healthy engines as well as 5 types of typical faulty engines occurred at the production line including operator’s remark. Auto Associative Kernel Regression (AAKR) model is used to separate the faulty engines from the healthy engine and it will also rate the engine health on a scale of 1 to 10 where 10 stands for excellent and smooth engine and 1 stands for severely faulty engine. Random Forest (RF) model is used to classify the type of fault in the engine. We have conducted a blind test with a set of healthy and faulty engines, we have achieved overall accuracy of 83.7% with true positive rate of 94 % and false positive rate of 25% for the health index model. Engine diagnostic model yields an overall accuracy of 75.9%