This master's thesis addresses the use of uncontrolled AI algorithms to improve the results of subjective classification and determining the validation interval. The improvement is based on the fact that the human brain is not able to objectively decide on the quality of a multidimensional problem.
It focuses on the pumps that will be used in water injection application in forced-induction petrol engines. This application requires a pump that would work in a wide spectrum of conditions, which is a challenge for the development, production, and assembly phases. While testing, the problem of product repeatability arose due to the manufacturing method and fine tolerances.
Pumps were tested on four operating points, which represent the points of effective operation in the application. The main outputs: flow rate, rotational speed of the electric motor, suction pressure in front of the pump, output pressure behind the pump, electrical output, voltage, current output, and phase voltage served as input data for algorithm classification. Due to the nature of these data, the most suitable algorithms proved to be Kohonen’s self-organizing map and the K-mean algorithm. As a results of algorithm classification, two main groups of pumps were identified. Pumps differ one from another, not in terms of flow rate depending on rotational speed of the electric motor, but in terms of efficiency, which deviates from subjective measurements.
|
