Identification of Mechanical Systems for Industrial Electric Drives : A review of algorithms for online parameter estimation and condition monitoring regardless of the amount of masses

Sammanfattning

System identification is frequently used for electric drives to estimate unknown parameters of the system being controlled, particularly the mechanical parameters. This is commonly done for two purposes: for actively tuning the speed control, and for diagnosis and monitoring. Offline identification plays an important role in industry, which implies that data is collected from the system during the commissioning process and system identification is performed outside of normal operation. Online identification presents advantages, it allows real-time monitoring and control tuning, and it has lower calculation and memory storage requirements. Then, it is of interest to perform the identification process in an online manner, for black-box systems and in a format that allows practical implementation in industrial environments. In this degree project, a review of methodologies is presented, and a three-step algorithm that can be used for online identification is designed and tested. For the first step, a non-parametric identification process using filtering is applied to identify the order of the system. A second step is to apply to an Output Error (OE) model a Recursive Least Squares (RLS) method, to identify the parameters of the transfer function governing the system. The third step is to extract the mechanical parameters from the identified transfer function. This degree project has relevant key results. It compares different signals that can be injected to excite the system to perform the identification process. Moreover, it provides a review of different parametric methodologies that were considered for this application. Furthermore, it makes a comparison of different recursive methodologies that can be used for parametric online identification, proposing the use of Recursive Weighted Least Squares (RWLS) method when injecting Chirp signals. Also, it performs a sensitivity analysis of the impact of the selected parameters for the identification process on the estimation results. Finally, future work is presented regarding the limitations that were encountered in this degree project.