Rotating machinery plays a vital role in industrial systems, in which unexpected mechanical faults during operation can lead to severe consequences. For fault prevention, many fault diagnostic methods based on vibration signals are available in the literature. However, vibration signals are obtained by using different types of sensors, which can cause sensor installation issues and damage the rotating machinery. In addition, this kind of data acquisition via vibration signal induces a large amount of signal noise during machine operation, which will challenge the later fault diagnosis. A recent fault detection method based on infrared thermography (IRT) for rotating machinery avoids these issues. However, the corresponding literature is limited by the fact that the characteristics of the manual design cannot fully characterize the fault, so that the diagnostic accuracy cannot exceed the diagnostic method based on the vibration signals. This paper introduces a popular image feature extraction method into the fault diagnosis of rotating machinery based on IRT for the first time. Capturing the IRT images of the rotating machinery in different states firstly, and then two popular feature extraction methods for IRT images, bag-of-visual-word (BoVW), and convolutional neural network (CNN), are tested in turn. Finally, the extracted features are classified to implement automatic fault diagnosis. The developed method is applied to analyze the experimental IRT images collected from bearings, and the results demonstrate that the developed method is more effective than the traditional methods based on vibration signals.

Electronics-rich analog systems are difficult to diagnose owing to their complex working mechanisms and the variability of the working environment. In recent years, deep learning has been gradually applied to the field of circuit system fault diagnosis because of its strong ability to mine the intrinsic characteristics of signals. However, the traditional deep learning method requires a lot of effort to achieve satisfactory results due to large number of parameters, complex models, slow training speed, and large datasets. The key factors for the success of traditional deep learning methods are layer-by-layer processing, feature transformation within the model, and sufficient model complexity. Deep forest (DF) is a new feature learning model that inherits the three characteristics of the traditional deep learning model but is that it is not based on neural network. It has fewer hyperparameters, a simpler model, faster training speed. In this article, an improved DF algorithm based on nonparametric predictive inference (NPI) is proposed, named NPIDF, which can better deal with small sample data. In two typical analog filter circuit fault diagnosis experiments, it is proved that DF and NPIDF achieve good diagnosis effect, and NPIDF performance is better, showing a greater advantage in small sample data.

Due to their short duration, concealability, and random occurrence, intermittent faults have become the most dangerous hazard in electronic circuit systems. However, existing intelligent diagnostic methods often struggle to provide substantial effectiveness for such objects. This article proposes a scheme to enhance the output signal of electronic circuits to make the features more intuitive and facilitate fault feature mining, and designs a scheme to mine tiny fault features from global signals. First, the circuit’s output signal undergoes an S-transform to obtain its time-frequency domain characteristics. Next, a sequence and excitation networks attention module is employed to allocate weights to different channels. Finally, the aforementioned output is used as input to the Swin transformer framework to thoroughly explore fault features. Three electronic circuits are used as experimental circuits to test the proposed method. The experiment shows that the proposed diagnostic method is fast and has an accuracy of over 97%. Therefore, the effectiveness of the designed strategy that includes multiple attention mechanisms in mining intermittent fault features in electronic circuit systems has been demonstrated.

The current popular machine learning-based fault diagnosis methods make it difficult to explain the diagnostic results, leading to low user trust in such diagnostic techniques. In this regard, this paper explores the study of the interpretability of intelligent fault diagnosis algorithms using the elevator of a fixed-wing unmanned aerial vehicle (UAV) as a diagnostic object. The Transformer model combines excellent modeling capability and efficient sequence data processing, is chosen to mine fault signal features to guarantee accurate diagnosis. Among the proposed interpretable fault diagnosis models, the local interpretable model-agnostic explanations (LIME) model is used to provide explicit interpretability for the decision-making process of the diagnosis model. In addition, a loss function called RDCE (reinforced diagnostic cross-entropy) Loss is designed to minimize the negative impact of different sample sizes for different fault types on the diagnostic performance. This loss function is designed to weigh the various types of faults to speed up the convergence of the model and improve the diagnostic accuracy. By comparing the proposed diagnostic strategy with other commonly used machine learning models, including long short term memory and recurrent neural network (RNN), the average diagnostic accuracy of the proposed diagnostic strategy is 99.97%, significantly better than that of the comparison algorithms. At the same time, this paper provides an in-depth interpretable analysis of the diagnostic process of the Transformer. The diagnostic process of the Transformer model gives the reasons for the diagnostic results from the point of view of the kind of features processed by the model. Based on this, the diagnostic model is simplified. After streamlining the number of features from 40 to 24 according to their importance, the diagnostic accuracy of the model is improved by 0.26%, and the diagnostic efficiency is improved. In addition, the proposed diagnostic strategy also shows significant advantages in terms of noise robustness.

The effectiveness and safety of an aircraft’s flight depend heavily on the flight control system. Since the attitude sensor is the weakest link, identifying its failure modes is crucial. To overcome the shortcomings of a single diagnosis model and a single input signal, this paper proposes a hybrid deep fault diagnosis model based on multi-data fusion. First, the normal and fault models of the sensor are established, and the residual timing signals of the sensor in different fault states are obtained. The frequency domain and timefrequency domain representations of the original timing signals are collected by means of fast Fourier transform and S-transform, and they are used as the input of the hybrid deep diagnosis model. The deep model is designed for the three inputs to mine the characteristics of the input data. These three deep features are concatenated and dimensionally reduced to obtain more comprehensive and representative features. Finally, the classifier is used to classify and obtain the diagnosis results. Through experiments, the advantages of the proposed method are verified by comparing it with several other methods.

The present paper proposes a multimodal deep learning model based on Recursive Fusion Feature Map (RFFM) and Transformer-TCN, to solve the problem of fault diagnosis of flying wing UAV actuators under complex working conditions. The challenges of data scarcity, high-dimensional nonlinear dynamic characteristics, and the difficulty of effectively fusing multimodal features are particularly relevant in this context. The recursive fusion feature map is used to extract high-order features from multi-modal time-series signals (position, velocity, acceleration, torque, current and voltage), and the Transformer is used to capture the long-term dependencies of the signals, while the TCN is employed to model short-term dynamic characteristics. This enables accurate classification and health assessment of the flying wing UAV actuator under normal, single fault (wear/jamming, dynamic lag, signal failure) and compound fault modes. In the simulation experiment, the six-modal time-series signal generated for the seven states was analysed. The experimental findings demonstrated that the proposed model attained a classification accuracy of 98.6% on the bal-anced dataset and 94.7% on the unbalanced dataset, with an F1 score exceeding 0.92 for each category. Concur-rently, the model’s resilience to complex fault modes was substantiated through a comparison of residual signals and an examination of time-frequency diagrams. A comparison of the model with traditional methods such as SVM, random forest and LSTM reveals significant advantages in key indicators such as average AUC value, diagnostic accuracy and classification stability (average AUC value exceeds 0.97). The research results demonstrate the effec-tiveness and applicability of this method in the complex fault diagnosis of flying wing UAV actuators and have important engineering application value and potential for promotion.