On Practical Aspects of Using RNNs for Fault Detection in Sparsely-labeled Multi-sensor Time Series

In this work, we attempt to address two practical limitations when using Recurrent Neural Networks (RNNs) as classifiers for fault detection using multi-sensor time series data: Firstly, there is a need to understand the classification decisions of RNNs. It is difficult for engineers to diagnose the faults when multiple sensors are being monitored at once. The faults detected by RNNs can be better understood if the sensors carrying the faulty signature are known. To achieve this, we propose a sensor relevance scoring (SRS) approach that scores each sensor based on its contribution to the classification decision by leveraging the hidden layer activations of RNNs. Secondly, lack of labeled training data due to infrequent faults (or otherwise) makes it difficult to train RNNs in a supervised manner. We pre-train an RNN on large unlabeled data via an autoencoder in an unsupervised manner, and then finetune the RNN for the fault detection task using small amount of labeled training data. Through experiments on a public gasoil heating loop dataset and a proprietary pump dataset, we demonstrate the efficacy of the proposed solutions, and show that i) SRS can help point to the sensors relevant for a fault, ii) large unlabeled data can be used to pre-train an RNNbased fault detector in an unsupervised manner in sparselylabeled scenarios, and iii) a purely unsupervised approach for fault detection (e.g. based on RNN-autoencoders) may not suffice when the number of sensors being monitored is large while the signature for fault is present in only a small subset of sensors.