Performance evaluation and prediction of rudders based on machine learning technology

Abstract

Rudders are the important components of aircrafts, missiles and ships, and their traditional test equipment is not intelligent enough, so we have to evaluate their performance by observing every parameter manually. This situation makes it impossible to test the rudders rapidly and in quantity. In this paper, we present a new application in the field of rudder test based on machine learning (ML) and describe new methods for performance evaluation and state prediction. The main topics are concentrated on prediction-oriented problems of multiple performance data mining and modeling: analysis and extraction of data feature, performance scoring based on regression algorithm and cross-validation, screening of defective products and fault location based on classification algorithm and accuracy evaluation. Besides, we propose a new optimized decision tree algorithm (SFLA-MWDT) which solves the common decision difficulty in tree models caused by low-precision decision and high-vote competition. Here, through ‘automatic acquisition + intelligent analysis’, we break through the shortcomings of traditional rudder testing methods and technical bottleneck of low parameter testing efficiency. This test method is applicable to those rudders that have already produced but not yet in use. Also, it provides guidance for the production and practice of rudders.