Research on operation condition assessment for escalators based on data augmentation methods: a case of hospital escalators

Abstract Escalators are common in hospitals that are highly trafficked public places. As an important part of equipment prognostics and health management, condition assessment plays an integral role. However, escalator condition assessment suffers from limited high-quality labeled data and unbalanced data categories. It creates challenges and barriers to escalator condition assessment. To this end, an auxiliary classifier (AC)-guided diffusion probabilistic model is proposed for escalator data augmentation and condition assessment. Firstly, real data is used to pre-train the AC. After the diffusion model forward process for noise addition, the noise data connected with category labels is fed into the symmetric network for noise prediction. Secondly, the diffusion model undergoes a reverse process for denoising. Among them, the samples generated at each step of the reverse denoising process are fed into the AC for screening, and the eligible samples are adopted to fine-tune the AC to effectively improve its generalization performance. The validity and feasibility of the proposed methodology are verified by the escalator’ operation data from different conditions. The novelty and necessity of the proposed method are demonstrated through multiple comparative experiments.