Preventive Maintenance Strategy for Air Conditioning Systems Based on Health Index Prediction

Constructing a health index (HI) that accurately describes the degradation process and accurately predicting the HI are two prerequisites for developing preventive maintenance (PM) strategy for air conditioning systems (ACS). Existing research primarily relies on single indicators when constructing the HI, which may fail to authentically reflect the system's health states. Moreover, an unreasonable HI can also render prediction model ineffective. To address the issue, this paper proposes a method that integrates multi-source data to construct and predict HI, which can help in conducting condition-based PM strategy. Specifically, a combination weighting method is employed to construct an HI using the set of monitoring parameters related to system condition, which provides a more reliable indicator for predicting health states. On this basis, the constructed HI is utilized as the label for the dataset, and long short-term memory (LSTM) neural networks are employed to predict the future HI of the system. Based on the predicted results of HI, corresponding maintenance measures are taken at different thresholds. The effectiveness of the proposed method is demonstrated through a practical example of ACS.