Data-driven fault detection approach for nonlinear three-tank system

Tee, Chin Jian (2021) Data-driven fault detection approach for nonlinear three-tank system. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

	Text (24 Pages) Data-driven fault detection approach for nonlinear three-tank system.pdf - Submitted Version Download (319kB)
	Text (Full text) Data-driven fault detection approach for nonlinear three-tank system.pdf - Submitted Version Restricted to Repository staff only Download (765kB)

Abstract

The data-driven is a progrress which in the formed of data instead of intuition by the people. It has capabilty to show of a process operating system and transfer the information to the moniring system in the form of data. Besides that, the fault detection is to indentify or detect the faults when it occurred. It also can be pointed the type of faults and its location of the three-tank system. The faults that may happen such as leakage fault, sensor failure, actuator failure, abrupt disturbance fault and so on. The main objectives of this research are to generate faulty and non-faulty data by using block diagram of MATLAB Simulink, to develop a principle componenet analysis (PCA) of data-driven fault detection method for three-tank system and to evaluate the performance of the method of data-driven fault detection by using confusion matric. The results for non-faulty event are 100% in accuraccy and specificity and below the threshold value which mean the the predicted non-faulty and actual non-faulty too. For the abrupt disturbance fault at tank 1, accuracy and sensitivity of the tank 1 dropped to 6.2% only. For leakage fault of tank 3, the accuracy and sensitivity are dropped to 97.2% and prediction is 100%. This shows that the predicted faulty and actual faulty too. For the sensor failure at tank 2, the accuracy, precision and sensitivity are 100% as the prediction made for this faulty is fully correct. Hence, the fault alarm will be ON when the system detected the occurance of the fault as the data is above the threshol value.

Actions (login required)

View Item