Modelling and FDI of Dynamic Discrete Time Systems Using a MLP with a New Sigmoidal Activation Function

In this paper we investigate the use of the multi-layer perceptron (MLP) for system modelling. A new sigmoidal activation function is introduced and the study is focused at the utilization of this function on a MLP that performs modelling of dynamic, discrete time systems. The role of the activation function in the training process is investigated analytically, and it is proven that the shape of the activation function and it's derivative can affect the training outcome. The method is simulated at a well known benchmark, namely the three tank system, and is incorporated in a Fault Detection and Identification (FDI) method, also applied and simulated at the three tank system. Finally, a comparison is made with an approach that utilizes local model neural networks for system modeling.     

Finding fault - fault diagnosis on the wheels of a mobile robot using local model neural networks

As mobile robots are mostly designed to act autonomously, procedures that detect and isolate faults on the various parts of a robot are essential. The most powerful approaches in fault detection and isolation (FDI) are those using a process model, where quantitative and qualitative knowledge-based models, databased models, or combinations thereof are applied. This article suggests a model-free approach to the solution of the fault detection problem. One way to deal with the absence of a mathematical model is to build a model from input-output data. In this article, local model networks (LMNs) are used for plant modeling. The key to fault detection and diagnosis is the creation of residual signals. Although the way these signals are formed varies, in all cases the residuals change their value accordingly with the presence of faults. To avoid false alarms, the residuals must be affected by factors unrelated to faults (like modeling errors) as little as possible. Change-detection algorithms are therefore used for reliable residual generation. These algorithms are designed to detect changes in signals that include noise or other types of disorders. The combination of local model networks for modeling and change-detection algorithms for residual creation provides an efficient method for fault detection and diagnosis. The method is applied on the wheels subsystem of a mobile robot.