Equipment failure is a typical problem, and it is difficult to detect it in the current technology-based industry. There are many different approaches to diagnostic reasoning such as Rule-based reasoning, model-based reasoning, and qualitative model-based reasoning in preventive maintenance. In this paper, we developed a model-based approach namely one-class Support Vector Machines (SVM) to detect the faults in the oil industry. Real data like pressure and temperature readings were collected from sensors which are placed at different parts of machines at each minute.  Also, failure occurrence of each machine is recorded. Before applying the model, as an initial step, the data (‘n’ hours just before the occurrence of a failure) is fine-tuned using Exponentially Weighted Moving Average (EWMA) method to remove the external noise factors which occurred while taking the readings. After fine-tuning the data, features like mean, median and peak values were extracted at the time interval of fifteen minutes. These features were also extracted using the correlation between sensor values and the information from the people who have domain knowledge on these machines. After this, we developed one-class SVM model on the data. We also built multiple one-class SVM models by ‘k’ fold cross validation which reduces overfitting. During this process, if the model indicates an outlier, it indicates that there is a chance of failure occurrence.