Gabriel Michau

Katharina Rombach, Gabriel Michau, Olga Fink

New operating conditions can result in a significant performance drop of fault diagnostics models due to the domain shift between the training and the testing data distributions. While several domain adaptation approaches have been proposed to overcome such domain shifts, their application is limited if the fault classes represented in the two domains are not the same. To enable a better transferability of the trained models between two different domains, particularly in setups where only the healthy data class is shared between the two domains, we propose a new framework for Partial and Open-Partial domain adaptation based on generating distinct fault signatures with a Wasserstein GAN. The main contribution of the proposed framework is the controlled synthetic fault data generation with two main distinct characteristics. Firstly, the proposed methodology enables to generate unobserved fault types in the target domain by having only access to the healthy samples in the target domain and faulty samples in the source domain. Secondly, the fault generation can be controlled to precisely generate distinct fault types and fault severity levels. The proposed method is especially suited in extreme domain adaption settings that are particularly relevant in the context of complex and safety-critical systems, where only one class is shared between the two domains. We evaluate the proposed framework on Partial as well as Open-Partial domain adaptation tasks on two bearing fault diagnostics case studies. Our experiments conducted in different label space settings showcase the versatility of the proposed framework. The proposed methodology provided superior results compared to other methods given large domain gaps.

Katharina Rombach, Gabriel Michau, Wilfried Bürzle et al.

Monitoring the health of complex industrial assets is crucial for safe and efficient operations. Health indicators that provide quantitative real-time insights into the health status of industrial assets over time serve as valuable tools for e.g. fault detection or prognostics. This study proposes a novel, versatile and unsupervised approach to learn health indicators using contrastive learning, where the operational time serves as a proxy for degradation. To highlight its versatility, the approach is evaluated on two tasks and case studies with different characteristics: wear assessment of milling machines and fault detection of railway wheels. Our results show that the proposed methodology effectively learns a health indicator that follows the wear of milling machines (0.97 correlation on average) and is suitable for fault detection in railway wheels (88.7% balanced accuracy). The conducted experiments demonstrate the versatility of the approach for various systems and health conditions.

Gabriel Michau, Gaetan Frusque, Olga Fink

High-Frequency (HF) signals are ubiquitous in the industrial world and are of great use for monitoring of industrial assets. Most deep learning tools are designed for inputs of fixed and/or very limited size and many successful applications of deep learning to the industrial context use as inputs extracted features, which is a manually and often arduously obtained compact representation of the original signal. In this paper, we propose a fully unsupervised deep learning framework that is able to extract a meaningful and sparse representation of raw HF signals. We embed in our architecture important properties of the fast discrete wavelet transformation (FDWT) such as (1) the cascade algorithm, (2) the conjugate quadrature filter property that links together the wavelet, the scaling and transposed filter functions, and (3) the coefficient denoising. Using deep learning, we make this architecture fully learnable: both the wavelet bases and the wavelet coefficient denoising are learnable. To achieve this objective, we propose a new activation function that performs a learnable hard-thresholding of the wavelet coefficients. With our framework, the denoising FDWT becomes a fully learnable unsupervised tool that does neither require any type of pre- nor post-processing, nor any prior knowledge on wavelet transform. We demonstrate the benefits of embedding all these properties on three machine-learning tasks performed on open source sound datasets. We perform an ablation study of the impact of each property on the performance of the architecture, achieve results well above baseline and outperform other state-of-the-art methods.

Gabriel Michau, Olga Fink

Anomaly Detectors are trained on healthy operating condition data and raise an alarm when the measured samples deviate from the training data distribution. This means that the samples used to train the model should be sufficient in quantity and representative of the healthy operating conditions. But for industrial systems subject to changing operating conditions, acquiring such comprehensive sets of samples requires a long collection period and delay the point at which the anomaly detector can be trained and put in operation. A solution to this problem is to perform unsupervised transfer learning (UTL), to transfer complementary data between different units. In the literature however, UTL aims at finding common structure between the datasets, to perform clustering or dimensionality reduction. Yet, the task of transferring and combining complementary training data has not been studied. Our proposed framework is designed to transfer complementary operating conditions between different units in a completely unsupervised way to train more robust anomaly detectors. It differs, thereby, from other unsupervised transfer learning works as it focuses on a one-class classification problem. The proposed methodology enables to detect anomalies in operating conditions only experienced by other units. The proposed end-to-end framework uses adversarial deep learning to ensure alignment of the different units' distributions. The framework introduces a new loss, inspired by a dimensionality reduction tool, to enforce the conservation of the inherent variability of each dataset, and uses state-of-the art once-class approach to detect anomalies. We demonstrate the benefit of the proposed framework using three open source datasets.

Gaetan Frusque, Gabriel Michau, Olga Fink

Acoustic monitoring for machine fault detection is a recent and expanding research path that has already provided promising results for industries. However, it is impossible to collect enough data to learn all types of faults from a machine. Thus, new algorithms, trained using data from healthy conditions only, were developed to perform unsupervised anomaly detection. A key issue in the development of these algorithms is the noise in the signals, as it impacts the anomaly detection performance. In this work, we propose a powerful data-driven and quasi non-parametric denoising strategy for spectral data based on a tensor decomposition: the Non-negative Canonical Polyadic (CP) decomposition. This method is particularly adapted for machine emitting stationary sound. We demonstrate in a case study, the Malfunctioning Industrial Machine Investigation and Inspection (MIMII) baseline, how the use of our denoising strategy leads to a sensible improvement of the unsupervised anomaly detection. Such approaches are capable to make sound-based monitoring of industrial processes more reliable.

Gabriel Rodriguez Garcia, Gabriel Michau, Herbert H. Einstein et al.

In tunnel construction projects, delays induce high costs. Thus, tunnel boring machines (TBM) operators aim for fast advance rates, without safety compromise, a difficult mission in uncertain ground environments. Finding the optimal control parameters based on the TBM sensors' measurements remains an open research question with large practical relevance. In this paper, we propose an intelligent decision support system developed in three steps. First past projects performances are evaluated with an optimality score, taking into account the advance rate and the working pressure safety. Then, a deep learning model learns the mapping between the TBM measurements and this optimality score. Last, in real application, the model provides incremental recommendations to improve the optimality, taking into account the current setting and measurements of the TBM. The proposed approach is evaluated on real micro-tunnelling project and demonstrates great promises for future projects.

Katharina Rombach, Gabriel Michau, Olga Fink

Mislabeled samples are ubiquitous in real-world datasets as rule-based or expert labeling is usually based on incorrect assumptions or subject to biased opinions. Neural networks can "memorize" these mislabeled samples and, as a result, exhibit poor generalization. This poses a critical issue in fault detection applications, where not only the training but also the validation datasets are prone to contain mislabeled samples. In this work, we propose a novel two-step framework for robust training with label noise. In the first step, we identify outliers (including the mislabeled samples) based on the update in the hypothesis space. In the second step, we propose different approaches to modifying the training data based on the identified outliers and a data augmentation technique. Contrary to previous approaches, we aim at finding a robust solution that is suitable for real-world applications, such as fault detection, where no clean, "noise-free" validation dataset is available. Under an approximate assumption about the upper limit of the label noise, we significantly improve the generalization ability of the model trained under massive label noise.

Gabriel Michau, Chi-Ching Hsu, Olga Fink

Partial discharge (PD) is a common indication of faults in power systems, such as generators, and cables. These PD can eventually result in costly repairs and substantial power outages. PD detection traditionally relies on hand-crafted features and domain expertise to identify very specific pulses in the electrical current, and the performance declines in the presence of noise or of superposed pulses. In this paper, we propose a novel end-to-end framework based on convolutional neural networks. The framework has two contributions. First, it does not require any feature extraction and enables robust PD detection. Second, we devise the pulse activation map. It provides interpretability of the results for the domain experts with the identification of the pulses that led to the detection of the PDs. The performance is evaluated on a public dataset for the detection of damaged power lines. An ablation study demonstrates the benefits of each part of the proposed framework.

Oliver Ammann, Gabriel Michau, Olga Fink

Recently, with the development of deep learning, end-to-end neural network architectures have been increasingly applied to condition monitoring signals. They have demonstrated superior performance for fault detection and classification, in particular using convolutional neural networks. Even more recently, an extension of the concept of convolution to the concept of kervolution has been proposed with some promising results in image classification tasks. In this paper, we explore the potential of kervolutional neural networks applied to time series data. We demonstrate that using a mixture of convolutional and kervolutional layers improves the model performance. The mixed model is first applied to a classification task in time series, as a benchmark dataset. Subsequently, the proposed mixed architecture is used to detect anomalies in time series data recorded by accelerometers on helicopters. We propose a residual-based anomaly detection approach using a temporal auto-encoder. We demonstrate that mixing kervolutional with convolutional layers in the encoder is more sensitive to variations in the input data and is able to detect anomalous time series in a better way.

Gabriel Rodriguez Garcia, Gabriel Michau, Mélanie Ducoffe et al.

The ability to detect anomalies in time series is considered highly valuable in numerous application domains. The sequential nature of time series objects is responsible for an additional feature complexity, ultimately requiring specialized approaches in order to solve the task. Essential characteristics of time series, situated outside the time domain, are often difficult to capture with state-of-the-art anomaly detection methods when no transformations have been applied to the time series. Inspired by the success of deep learning methods in computer vision, several studies have proposed transforming time series into image-like representations, used as inputs for deep learning models, and have led to very promising results in classification tasks. In this paper, we first review the signal to image encoding approaches found in the literature. Second, we propose modifications to some of their original formulations to make them more robust to the variability in large datasets. Third, we compare them on the basis of a common unsupervised task to demonstrate how the choice of the encoding can impact the results when used in the same deep learning architecture. We thus provide a comparison between six encoding algorithms with and without the proposed modifications. The selected encoding methods are Gramian Angular Field, Markov Transition Field, recurrence plot, grey scale encoding, spectrogram, and scalogram. We also compare the results achieved with the raw signal used as input for another deep learning model. We demonstrate that some encodings have a competitive advantage and might be worth considering within a deep learning framework. The comparison is performed on a dataset collected and released by Airbus SAS, containing highly complex vibration measurements from real helicopter flight tests. The different encodings provide competitive results for anomaly detection.

Qin Wang, Gabriel Michau, Olga Fink

Domain adaptation aims at improving model performance by leveraging the learned knowledge in the source domain and transferring it to the target domain. Recently, domain adversarial methods have been particularly successful in alleviating the distribution shift between the source and the target domains. However, these methods assume an identical label space between the two domains. This assumption imposes a significant limitation for real applications since the target training set may not contain the complete set of classes. We demonstrate in this paper that the performance of domain adversarial methods can be vulnerable to an incomplete target label space during training. To overcome this issue, we propose a two-stage unilateral alignment approach. The proposed methodology makes use of the inter-class relationships of the source domain and aligns unilaterally the target to the source domain. The benefits of the proposed methodology are first evaluated on the MNIST$\rightarrow$MNIST-M adaptation task. The proposed methodology is also evaluated on a fault diagnosis task, where the problem of missing fault types in the target training dataset is common in practice. Both experiments demonstrate the effectiveness of the proposed methodology.

Gabriel Michau, Olga Fink

The failure of a complex and safety critical industrial asset can have extremely high consequences. Close monitoring for early detection of abnormal system conditions is therefore required. Data-driven solutions to this problem have been limited for two reasons: First, safety critical assets are designed and maintained to be highly reliable and faults are rare. Fault detection can thus not be solved with supervised learning. Second, complex industrial systems usually have long lifetime during which they face very different operating conditions. In the early life of the system, the collected data is probably not representative of future operating conditions, making it challenging to train a robust model. In this paper, we propose a methodology to monitor the systems in their early life. To do so, we enhance the training dataset with other units from a fleet, for which longer observations are available. Since each unit has its own specificity, we propose to extract features made independent of their origin by three unsupervised feature alignment techniques. First, using a variational encoder, we impose a shared probabilistic encoder/decoder for both units. Second, we introduce a new loss designed to conserve inter-point spacial relationships between the input and the learned features. Last, we propose to train in an adversarial manner a discriminator on the origin of the features. Once aligned, the features are fed to a one-class classifier to monitor the health of the system. By exploring the different combinations of the proposed alignment strategies, and by testing them on a real case study, a fleet composed of 112 power plants operated in different geographical locations and under very different operating regimes, we demonstrate that this alignment is necessary and beneficial.

Gabriel Michau, Olga Fink

Training data-driven approaches for complex industrial system health monitoring is challenging. When data on faulty conditions are rare or not available, the training has to be performed in a unsupervised manner. In addition, when the observation period, used for training, is kept short, to be able to monitor the system in its early life, the training data might not be representative of all the system normal operating conditions. In this paper, we propose five approaches to perform fault detection in such context. Two approaches rely on the data from the unit to be monitored only: the baseline is trained on the early life of the unit. An incremental learning procedure tries to learn new operating conditions as they arise. Three other approaches take advantage of data from other similar units within a fleet. In two cases, units are directly compared to each other with similarity measures, and the data from similar units are combined in the training set. We propose, in the third case, a new deep-learning methodology to perform, first, a feature alignment of different units with an Unsupervised Feature Alignment Network (UFAN). Then, features of both units are combined in the training set of the fault detection neural network. The approaches are tested on a fleet comprising 112 units, observed over one year of data. All approaches proposed here are an improvement to the baseline, trained with two months of data only. As units in the fleet are found to be very dissimilar, the new architecture UFAN, that aligns units in the feature space, is outperforming others.

Qin Wang, Gabriel Michau, Olga Fink

Thanks to digitization of industrial assets in fleets, the ambitious goal of transferring fault diagnosis models fromone machine to the other has raised great interest. Solving these domain adaptive transfer learning tasks has the potential to save large efforts on manually labeling data and modifying models for new machines in the same fleet. Although data-driven methods have shown great potential in fault diagnosis applications, their ability to generalize on new machines and new working conditions are limited because of their tendency to overfit to the training set in reality. One promising solution to this problem is to use domain adaptation techniques. It aims to improve model performance on the target new machine. Inspired by its successful implementation in computer vision, we introduced Domain-Adversarial Neural Networks (DANN) to our context, along with two other popular methods existing in previous fault diagnosis research. We then carefully justify the applicability of these methods in realistic fault diagnosis settings, and offer a unified experimental protocol for a fair comparison between domain adaptation methods for fault diagnosis problems.

Gabriel Michau, Yang Hu, Thomas Palmé et al.

Complex industrial systems are continuously monitored by a large number of heterogeneous sensors. The diversity of their operating conditions and the possible fault types make it impossible to collect enough data for learning all the possible fault patterns. The paper proposes an integrated automatic unsupervised feature learning and one-class classification for fault detection that uses data on healthy conditions only for its training. The approach is based on stacked Extreme Learning Machines (namely Hierarchical, or HELM) and comprises an autoencoder, performing unsupervised feature learning, stacked with a one-class classifier monitoring the distance of the test data to the training healthy class, thereby assessing the health of the system. This study provides a comprehensive evaluation of HELM fault detection capability compared to other machine learning approaches, such as stand-alone one-class classifiers (ELM and SVM), these same one-class classifiers combined with traditional dimensionality reduction methods (PCA) and a Deep Belief Network. The performance is first evaluated on a synthetic dataset that encompasses typical characteristics of condition monitoring data. Subsequently, the approach is evaluated on a real case study of a power plant fault. The proposed algorithm for fault detection, combining feature learning with the one-class classifier, demonstrates a better performance, particularly in cases where condition monitoring data contain several non-informative signals.