Novel double layer BiLSTM minor soft fault detection for sensors in air-conditioning system with KPCA reducing dimensions

Abstract

Abstract The initial stages of sensor faults of air-conditioning systems are not easy to detect. They may affect the normal operation of air-conditioning systems, resulting in loss of energy and a quality reduction of the indoor environment. To accurately detect minor soft faults of sensors, this study presents a novel method of combining kernel principal component analysis and double layer bidirectional long short-term memory (KPCA-DL-BiLSTM). Firstly, kernel principal component analysis (KPCA) extracts the principal components of related variables (chilled water valve opening, supply air temperature , fresh air temperature, fresh air humidity, return air temperature , and return air humidity) of the air-conditioning system and reduces the dimensionality of features. Subsequently, the time characteristics of the data are converted into sequences with sliding windows, which are used as the input of double layer bidirectional long short-term memory (DL-BiLSTM). Finally, minor soft faults of sensors can be detected by the residual which is generated by comparing the output of DL-BiLSTM with the actual value from the supply air temperature sensor. The key contribution of this paper is to study the time dependence of sensor faults to improve the detection rate of minor faults. The experimental results show that the detection accuracy of KPCA-DL-BiLSTM was 43% higher than that of KPCA and 18.33% higher than that of the long short-term memory (LSTM) under a 10% drift deviation fault. It can be seen from the results that KPCA-DL-BiLSTM had better fault detection accuracy and stability for minor soft faults, especially for drift deviation faults.