Abstract :A model has been developed for the health assessment of a chlorine compressor that is capable of constructing Health Indicators (HI). This model is an LSTM Autoencoder, which works as follows: an LSTM-based encoder maps a multivariate input sequence to a fixed-dimensional vector representation. The decoder, another LSTM network, uses this vector representation to produce the target sequence. The loss between the reconstructed sequence and the input sequence forms the basis for the HI. The model is trained exclusively with healthy data. The LSTM-AE is designed to reconstruct the measurements independently of the operating conditions, enhancing the model’s robustness in varying operational contexts. Health assessment is conducted by using the constructed HI to classify the health state of the compressor. The first instance of the compressor being marked as unhealthy triggers the start of fault identification. The proposed method is tested on real-world data through a case study on chlorine compressors, resulting in an identification accuracy of 73% and a precision of 56% for the three considered failure mechanisms.