Improving Predictive Maintenance Performance with Limited Data under Different Operating Conditions with Transfer Learning

This paper aims to provide an effective solution to the problem of data constraints that are frequently encountered in artificial intelligence applications. A combination of deep learning techniques is proposed to overcome the challenges related to data access and quality in the remaining useful life (RUL) estimation of industrial equipment. Using data obtained from equipment operating in different operating conditions, the research aims to perform RUL prediction of equipment operating in another operational context with the highest accuracy at the lowest cost. The applied methodology includes various Transfer Learning algorithms and domain adaptation techniques. Experimental results show that the proposed approach achieves a significant improvement in RUL prediction performance by 60% on average compared to the scenario where the proposed methods are not applied. The work contributes to the fields of predictive maintenance and asset management in industrial systems, offering the potential to optimize reliability and cost where data constraints are common like defence industry.