Self-Diagnosis Features for Estimation Models

Nowadays, developing estimation algorithms that can provide an accurate and optimal estimation of the targeted output remains to be a challenging topic for both research and industrial areas. Despite the numerous numbers of research done on the topic of improving the efficiency and accuracy of different estimation algorithms, there is always a margin of failure that can be caused by the infinite size of the input space where it can make the algorithm unable to estimate the correct output. This problem is specifically evident during runtime since many uncertainties will rise that are caused by different working conditions and it is impossible to know if our algorithm is providing accurate estimation or not. To address this problem, this academic work and in collaboration with Brain Technologies, proposes new state-of-the-art methodologies titled the self-diagnosis features. It provides the estimation algorithm the ability to carry out an on-line self-assessment of their performance. In addition, it can be equipped to different types of applications that rely on the estimation algorithms. These new methodologies are based on the unsupervised machine learning algorithm of the data clustering. The main idea is to divide the input space into multiple clusters where each cluster correspond to a performance accuracy range. In this way, the expected performance of the estimation algorithm for the new data arriving during runtime can be estimated by the Euclidean distance from the cluster’s centroids. The approach adopted in this thesis work is firstly to have a dissertation of self-diagnosis features, study the state of art and present the theory for the mentioned methodologies. Lastly, two studies were performed on two applications to demonstrate the practicability of the approach, which they are: •??The Estimation of the state of charge (SOC) of the battery management system •??The Estimation of the state of health (SOH) of the CNC machine Following These studies, a conclusion shall be provided to prove the feasibility of the approach and provide a final review on the methodologies.