Incipient Fault Detection Based on Kolmogorov-Smironv Test

In chemical processes, there are both nonlinearity and slow dynamics among process variables. Meanwhile, at the beginning of faults, slight variation may not prompt the significant parametric changes. Traditional fault detection methods, based on mean and variation shifts, often fail to achieve satisfactory performance. To improve the detection of incipient quality-related faults, a novel orthonormal slow feature-based quality-related data decomposition network with Jensen-Shannon divergence (OSF-JS) fault detection method is proposed. The orthonormal subspace analysis (OSA) is embedded with Siamese to extract orthonormal slow features (OSFs) from decomposed subspaces. Enhancing the detection of incipient faults, JS divergence combined with a sliding window algorithm is introduced to construct statistics. The effectiveness of OSF-JS is validated for a wastewater treatment process and a real-world chemical process. Results show that the proposed OSF-JS method not only improves the fault detection rate (FDR) and reduces false alarm rate (FAR) but also identifies whether faults occur in the quality-related component.

The increasing complexity of power systems as well as the rapid expansion of renewable energy resources and deregulation of electricity industry lead to continual demand for operating power systems at higher efficiencies. The modern power systems are pushed to operate close to their secure limits. When subjected to a severe contingency, the stressed power systems may lose its dynamic security, resulting in catastrophic consequences such as cascading failure and even wide-spread blackout. Under this new reality, online and real-time power system stability assessments have great impacts on avoiding blackout events. The traditional stability methods are mainly based on time-domain simulation, which are not fast enough to satisfy the online and real-time application requirements. In recent years, with the development of advanced real-time system monitoring and communication techniques, the data-driven methods have been identified as powerful tools to help power systems counteract large disturbance and avoid instability. To solve above threats, this Ph.D. research aims to develop data-analytics methodologies from power grid measurement data to recognize the system stability, covering both pre-fault dynamic security assessment (DSA) and post-fault real-time short-term voltage stability (STVS) assessment areas under some extreme scenarios and practical issues. Specific objectives of this Ph.D. research are to develop an intelligent system for the future smart grid. Based on several kinds of promising advanced data-analytics technologies, including randomized learning, incremental broad learning (BL), transfer learning (TL), adversarial machine learning (ML), and generative adversarial network (GAN), etc, an intelligent system will be developed to interpret the power grid stability characteristics, and to support stability-constrained operation control actions. Hence, it can make power systems achieve seven goals, including extensibility, efficiency, robustness, security, personalization, reliability, and effectiveness. The outcomes of this Ph.D. research will enable a more secure, intelligent, and knowledge-transparent future of the next-generation smart grid. The main contributions through this Ph.D. research are as follows: Firstly, the temporal-adaptive data-driven methods are proposed based on optimized hybrid ensemble learning mechanism for online STVS assessment. Compared to the traditional data-driven stability methods, the proposed methods can balance the tradeoff between the stability accuracy and speed and select the most suitable Pareto set. Secondly, in the area of DSA, an online updating DSA method is developed by using incremental BL technique. The incremental BL method only needs to calculate the pseudoinverse of new state matrix without the computations of the whole output weights matrix. In this way, it does not need re-training the whole model to continuously maintain and improve the online DSA accuracy. Besides, the proposed method can achieve under three different increment scenarios, including increment of enhancement hidden nodes, features, and new training instances. Thirdly, in the area of missing phasor measurement unit (PMU) data, two different fully robust data-driven stability methods are developed, respectively for pre-fault DSA and post-fault STVS assessment. The former one is based on GAN without observability-constrained feature subsets to ensure its availability under any PMU missing scenario. The latter one is based on super-resolution perception and incremental BL, which can achieve both satisfactory stability accuracy and speed. Fourthly, in order to accurately quantify the robustness of the ML-based model under adversarial perturbations on safety-critical power system stability problem, an adversarial robustness verification method for data-driven stability models is proposed. Besides, a model-free and attack-independent robust index are defined to evaluate the ability of the ML-based stability models against adversarial examples, which can be used to select the candidate model and provide formal robustness guarantee for stability application in practice. Rigorous mathematic proof is provided for the robust index under both linear and non-linear binary scenarios. Moreover, the proposed robust index can be used to mitigate the adversarial examples via adversarial training. Finally, three different TL-based DSA methods are proposed to use one known fault to evaluate other related but different unknown faults. The first two methods aim to reduce the distribution differences between the trained data and unknown data with the complete unlearned faults. The third one considers the different faults with incomplete data scenario. Through applying the adversarial training process, the DSA model trained by one fault can work for a different but related fault by confusing domain discriminator. With the property of feature extractor network, the proposed integrated TL-based method can solve the missing data issue for the unknown faults. All the methodologies have been tested on benchmark power systems to verify their effectiveness, and comparative and comparative studies with existing methods in the literature are conducted where applicable.

In this paper, we present a synergistic approach to startup fault detection and diagnosis in gas turbine engines. The method employs statistics, signal processing and soft computing techniques in a complementary manner to address fault detection at transient conditions. Traditional turbine engine fault detection and diagnosis methods are based on engine data collected at steady-state conditions. However, insipient faults are difficult to diagnose using steady-state engine date; only engine faults that are fairly developed can be detected using conventional methods. Because incipient engine component faults are often manifested in the engine startup characteristics, we present a method of characterizing the engine transient startup using the following steps: engine sensor data during startup are recorded in time series format. The sensor profiles corresponding to "good" and "bad" engine startups are sampled using the bootstrap technique. A feature vector is extracted from these data using principal component analysis (PCA). Then, several important discriminating features are distilled from the feature vector. The features obtained from this step are then classified using neural-network-based methods. The "leave-one-out" approach to cross-validation is applied to obtain an objective evaluation of the neural network training. The proposed fault detection and diagnosis method is evaluated using actual engine startup data and the results are presented.

An interpretable data-driven method for degradation prediction of proton exchange membrane fuel cells based on temporal fusion transformer and covariates

Accurate prediction of endpoint carbon at the dynamic control stage in the converter is crucial for achieving smelting targets. Currently, there are two main methods for converter endpoint prediction: the data-driven method and the mechanism-based method. Data-driven methods exhibit high accuracy but are vulnerable to data quality variations and lack interpretability. Mechanism-based methods provide great interpretability but face challenges in precisely identifying key parameters in the mechanism formula. Inspired by the design concept of physics-informed neural networks (PINNs), an integrated data-driven and mechanism-based method for endpoint carbon prediction in BOF (dmPINNs, data-driven and mechanism-based physics-informed neural networks) is proposed, which has four parts: feature extraction, mechanism-based calculation, data-driven prediction, and integrated prediction. We identify key parameters of the mechanism formula through the neural network to obtain the specified formula for each heat and supervise the training process of the neural network through the mechanism formula to ensure interpretability. Experimental results show that, within the ±0.012% error range, the hit rate of endpoint carbon content using dmPINNs improved by 5.23% compared with the traditional data-driven method and has greater robustness with the supervision of the mechanism formula.

Slope stability prediction is a crucial aspect of landslide disaster prevention and control. Traditional methods are physics-based or data-driven, which are limited by their respective reliance on geotechnical engineering expertise and scarce real-world slope data. This study developed a slope stability prediction model that integrates physics and data through an innovative transfer learning (PD-TL) framework. Four TrAdaBoost transfer learning models were developed using Support Vector Classification (SVC), Logistic Regression (LR), Decision Tree (DT), and Random Forest (RF) as base learners. At the data level, the finite element analysis cases significantly expanded the sample size to resolve issues with data scarcity and model performance limited by sample size. At the model level, most traditional data-driven methods assume that training and test datasets follow the same distribution; in contrast, the proposed model relaxes this assumption and thus offers a promising solution to the scarcity of training data. At the knowledge level, the model incorporates geotechnical engineering expertise to enhance performance by leveraging field-specific insights and principles. Among the TrAdaBoost (-SVC, LR, DT, RF) models tested, TrAdaBoost-SVC showed superior slope stability prediction performance (ACC = 0.878, Precision = 0.794, Recall = 0.931, F1 = 0.857, AUC = 0.928, TNR = 84%). Its superiority demonstrates the value of integrating physics and data into a double-driven framework, which outperforms traditional data-driven and ensemble learning methods. The results also may represent suggestions for future slope stability analysis and broader engineering applications.

This study delves into the property submarket in Kuala Lumpur and Selangor, Malaysia. The submarket is anticipated to be simple, uniform, and dense, making it highly influenced by neighbouring properties. However, traditional data-driven methods that overlook spatial contiguity disregard this density condition. To tackle this problem, the study investigates spatially constrained data-driven methods utilizing Principal Component Analysis (PCA) and cluster analysis. The findings reveal that spatially constrained methods outperform traditional methods by minimizing errors and enhancing model fit. Specifically, the two-step cluster method and k-means cluster method reduce errors by 6.96% and 7.22%, respectively, but at the cost of model fit by 11.23% and 13.94%. Conversely, the spatial k-means and spatial agglomerative hierarchical cluster methods reduce errors by 8.68% and 8.17%, respectively, while improving model fit by 7.1% and 6.35%. Hence, the study concludes that spatially constrained data-driven methods are more effective in differentiating commercial property submarkets than traditional methods.

The data-driven methods based multivariate regression have become popular in the area of fault detection due to the development of the computer technique. However, some traditional data-driven methods only consider the statical operating environment that the dynamic relationship in the variables will be ignored to bring some false detection results. In this study, an approach called the dynamic fault detection (DFD) is proposed to solve dynamic behavior under the nonlinear case. From the view of the best KPIs, the proposed method divides the variables into two orthogonal subspaces by the improved kernel principal component regression to judge whether the happened fault is relevant to KPIs or not. Finally, in the numerical simulation, the effectiveness of the DFD approach is demonstrated by comparing it with three nonlinear methods.

Early detection of incipient faults is of vital importance to reducing maintenance costs, saving energy, and enhancing occupant comfort in buildings. Popular supervised learning models such as deep neural networks are considered promising due to their ability to directly learn from labeled fault data; however, it is known that the performance of supervised learning approaches highly relies on the availability and quality of labeled training data. In Fault Detection and Diagnosis (FDD) applications, the lack of labeled incipient fault data has posed a major challenge to applying these supervised learning techniques to commercial buildings. To overcome this challenge, this paper proposes using Monte Carlo dropout (MC-dropout) to enhance the supervised learning pipeline, so that the resulting neural network is able to detect and diagnose unseen incipient fault examples. We also examine the proposed MC-dropout method on the RP-1043 dataset to demonstrate its effectiveness in indicating the most likely incipient fault types.

The incipient fault detection of a complex industrial process is a challenging problem for traditional dynamic detection methods. Traditional dynamic detection methods usually decouple the correlations among the variables and dynamic correlations simultaneously, which makes the two types of correlations mixed and may lead to performance deterioration in long-sequence dynamic detection. Some incipient faults may not change the amplitudes of process variables but change the long-sequence dynamic features. Based on the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$T^{2}$</tex-math> </inline-formula> statistic and matrix multiplication transformation ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$T^{2}$</tex-math> </inline-formula> S-MMT), traditional dynamic detection methods can detect many faults effectively. However, the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$T^{2}$</tex-math> </inline-formula> S-MMT can not effectively detect some incipient faults due to the above two types of correlations mixed. In order to overcome the shortcomings of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$T^{2}$</tex-math> </inline-formula> S-MMT and improve the detection ability of some incipient faults, this paper proposes a fault detection method based on the dynamic k-nearest neighbor model and Dual Control Chart (DKNN-DCC), which can improve the incipient fault detection performance by using long-sequence dynamic detection. The proposed method is verified by the Tennessee Eastman (TE) process and the continuously stirred tank reactor (CSTR) process. The experimental results show the effectiveness of the proposed method in incipient fault detection compared with traditional dynamic detection methods. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Note to Practitioners</i> —This paper presents a novel incipient fault detection method, which directly mines the long-sequence dynamic abnormal information from the process variable and overcomes the problem of some abnormal information being submerged in the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$T^{2}$</tex-math> </inline-formula> statistic calculated based on the matrix multiplication transformation. The proposed method can detect incipient faults that are not easily detected by some traditional methods and can help operators find the abnormal and avoid more serious losses. The structure of the proposed method jumps out of the frameworks of traditional dynamic detection methods, which is feasible to apply to different stable industrial processes.

A hybrid-driven probabilistic state space model for tool wear monitoring

In the field of industrial maintenance, early detection of gearbox faults is crucial, yet challenging, especially under varying and unseen operating conditions and types of damage. Traditional gear fault detection methods are typically designed for known operating conditions. However, in real industrial scenarios, gear fault detection is always an open-set detection, in which working conditions are commonly unseen. To overcome these limitations, this study introduces the ResNetAutoencoder Fault Detection (RAFD) method. RAFD integrates a short-time Fourier transform with a pre-trained ResNet-18 model for feature extraction and an autoencoder for feature reconstruction, trained on normal operational data. Moreover, a similarity loss is introduced to minimize the distribution gap of different working conditions. The effectiveness of the method is evaluated against traditional autoencoder and one-class support vector machine across various operational conditions, with performance metrics including AUROC, precision, recall, and F1 -score. The results demonstrate RAFD's superior performance in both in-distribution and out-of-distribution tasks, underscoring its robustness and reliability in detecting normal and faulty gearbox states.

Passive sensing applications are limited by their inability to determine who is using a device, a critical concern in child mobile device use research, where devices are often shared between siblings or between a child and their parent. Our previous work leveraged behavioral biometrics to identify a target child user; however, it is unknown what type of training data is necessary for optimal model performance. This study evaluated model performance across different characteristics of training data. Thirty-six children (11.3 ± 0.9 years, 56% female) self-selected a video or a game on iPads for 10 min while laying and for another 5 min while sitting. The SensorLog application captured iPad accelerometer and gyroscope data while the child interacted with the device. Machine learning algorithms including Neural Network (NN), Random Forest (RF), k-Nearest Neighbors (k-NN), and SwipeFormer were applied to determine the most important aspects of training data to optimize model performance. The aspects of training data evaluated included (1) varying the length (i.e., seconds of training data), (2) varying the user position (i.e., sitting, laying), and (3) varying the time proximity between training and testing data. F1 score was used to evaluate model performance. The SwipeFormer F1 scores were lowest when the training data was further from the test data (0 when training data was 11 min away from test data) and highest when training data was close to test data (0.91 when training data was the minute preceding test data). The SwipeFormer F1 scores were highest when predicting the user laying from laying (0.97) and sitting from sitting (0.94), and lowest when predicting the user sitting from laying (0) and laying from sitting (0). The length of training data had little impact on performance, with a SwipeFormer F1 score of 0.91 when training on one minute of data and a SwipeFormer F1 score of 0.94 when training on twelve minutes of data. Because researchers would likely be predicting users at different timepoints than their training data, research should focus on improving model performance for identifying users independent of time proximity for training and test data.

A centrifugal pump is one type of pumps that widely used in industries. Its mechanism which creates pressure changes may cause cavitation. Cavitation phenomenon that is not properly maintained may results fatal breakdown leading to high economic losses. Therefore, research is needed to find and develop a method that can detect early cavitation phenomena and identify it at several levels as well. This paper presents a method that can detect cavitation by monitoring the vibrations level of the pump based on statistical analysis of time domain and Principal Component Analysis (PCA). Vibration data is collected, trained and tested for each cavitation level. Training data is normalized and trained for each cavitation level using PCA which produces data loading matrix. The loading matrix is then multiplied by the testing data which gives a score matrix used to classify cavitation level of the centrifugal pump. The result shows that the method of domain-based PCA is successful in transforming the original data of 7 statistical parameters to 7 principal components (PC) with maximum variant. Three PCs gives 93.68% variants which can clearly identify and classify the differences between normal, early, intermediate and fully developed cavitation in the centrifugal pumps.

Voltage sags are often manifested as the permanent or incipient faults occurred in the power system because of equipment malfunctions or failures. The incipient faults which are originally self-cleaning faults may repeatedly occur and gradually develop to a permanent fault after its first occurrence. The incipient fault detection is considered as a crucial task in predictive maintenance for power equipment such as transformers, circuit breakers, and underground cables. This paper proposes a hybrid method for incipient faults detection and classification. The proposed method firstly adopts two extraction methods and two feature measures to obtain seven peculiar features from voltage waveforms of abnormal phases recorded by power quality monitors at substations in a transmission network. Then, a feature selection method and the support vector machine combined with particle swarm optimization are applied to classify various types of incipient faults. Test results show that the proposed method contributes relatively accurate classification of incipient faults and can be employed as a useful tool for condition monitoring of major power equipment in the transmission network.