Search

This paper proposes a novel and reliable leak-detection method for pipeline systems based on acoustic emission (AE) signals. The proposed method analyzes signals from two AE sensors installed on the pipeline to detect leaks located between these two sensors. Firstly, the raw AE signals are preprocessed using empirical mode decomposition. The time difference of arrival (TDOA) is then extracted as a statistical feature of the two AE signals. The state of the pipeline (leakage/normal) is determined through comparing the statistical distribution of the TDOA of the current state with the prior normal state. Specifically, the two-sample Kolmogorov-Smirnov (K-S) test is applied to compare the statistical distribution of the TDOA feature for leak and non-leak scenarios. The K-S test statistic value in this context functions as a leakage indicator. A new criterion called leak sensitivity is introduced to evaluate and compare the performance of leak detection methods. Extensive experiments were conducted using an industrial pipeline system, and the results demonstrate the excellence of the proposed method in leak detection. Compared to traditional feature-based indicators, our approach achieves a significantly higher performance in leak detection.

In this paper, an approach to perform leak state detection and size identification for industrial fluid pipelines with an acoustic emission (AE) activity intensity index curve (AIIC), using b-value and a random forest (RF), is proposed. Initially, the b-value was calculated from pre-processed AE data, which was then utilized to construct AIICs. The AIIC presents a robust description of AE intensity, especially for detecting the leaking state, even with the complication of the multi-source problem of AE events (AEEs), in which there are other sources, rather than just leaking, contributing to the AE activity. In addition, it shows the capability to not just discriminate between normal and leaking states, but also to distinguish different leak sizes. To calculate the probability of a state change from normal condition to leakage, a changepoint detection method, using a Bayesian ensemble, was utilized. After the leak is detected, size identification is performed by feeding the AIIC to the RF. The experimental results were compared with two cutting-edge methods under different scenarios with various pressure levels and leak sizes, and the proposed method outperformed both the earlier algorithms in terms of accuracy.

Trip perturbations are proposed to be a leading cause of falls in older adults. To prevent trip-falls, trip-related fall risk should be assessed and subsequent task-specific interventions improving recovery skills from forward balance loss should be provided to the individuals at risk of trip-fall. Therefore, this study aimed to develop trip-related fall risk prediction models from one's regular gait pattern using machine-learning approaches. A total of 298 older adults (≥60 years) who experienced a novel obstacle-induced trip perturbation in the laboratory were included in this study. Their trip outcomes were classified into three classes: no-falls ( n = 192), falls with lowering strategy (L-fall, n = 84), and falls with elevating strategy (E-fall, n = 22). A total of 40 gait characteristics, which could potentially affect trip outcomes, were calculated in the regular walking trial before the trip trial. The top 50% of features ( n = 20) were selected to train the prediction models using a relief-based feature selection algorithm, and an ensemble classification model was selected and trained with different numbers of features (1-20). A ten-times five-fold stratified method was utilized for cross-validation. Our results suggested that the trained models with different feature numbers showed an overall accuracy between 67% and 89% at the default cutoff and between 70% and 94% at the optimal cutoff. The prediction accuracy roughly increased along with the number of features. Among all the models, the one with 17 features could be considered the best model with the highest AUC of 0.96, and the model with 8 features could be considered the optimal model, which had a comparable AUC of 0.93 and fewer features. This study revealed that gait characteristics in regular walking could accurately predict the trip-related fall risk for healthy older adults, and the developed models could be a helpful assessment tool to identify the individuals at risk of trip-falls.

In this study, a scheme for leak localization on a cylinder tank bottom using acoustic emission (AE) is proposed. This approach provides a means of early failure detection, thus reducing financial damage and hazards to the environment and users. The scheme starts with the hit detection process using a constant false alarm rate (CFAR) and a fixed thresholding method for a time of arrival (TOA) and an end-time determination. The detected hits are then investigated to group those originating from the same AE source together by enforcing an event definition and a similarity score. Afterwards, these newly grouped hits are processed by a time difference of arrival (TDOA) to find the locations of the events. Since the locations of the events alone do not pinpoint the leak location, a data density analysis using a Voronoi diagram is employed to find the area with the highest possibility of a leak's existence. The proposed method was validated using the Hsu-Nielsen test on a cylinder tank bottom under a one-failed-sensor scenario, which returned a highly accurate result across multiple test locations.

Accurate segmentation of drivable areas and road obstacles is critical for autonomous mobile robots to navigate safely in indoor and outdoor environments. With the fast advancement of deep learning, mobile robots may now perform autonomous navigation based on what they learned in the learning phase. On the other hand, existing techniques often have low performance when confronted with complex situations since unfamiliar objects are not included in the training dataset. Additionally, the use of a large amount of labeled data is generally essential for training deep neural networks to achieve good performance, which is time-consuming and labor-intensive. Thus, this paper presents a solution to these issues by proposing a self-supervised learning method for the drivable areas and road anomaly segmentation. First, we propose the Automatic Generating Segmentation Label (AGSL) framework, which is an efficient system automatically generating segmentation labels for drivable areas and road anomalies by finding dissimilarities between the input and resynthesized image and localizing obstacles in the disparity map. Then, we train RGB-D datasets with a semantic segmentation network using self-generated ground truth labels derived from our method (AGSL labels) to get the pre-trained model. The results showed that our AGSL achieved high performance in labeling evaluation, and the pre-trained model also obtains certain confidence in real-time segmentation application on mobile robots.

This paper proposes a new technique for the construction of a concrete-beam health indicator based on the Kullback-Leibler divergence (KLD) and deep learning. Health indicator (HI) construction is a vital part of remaining useful lifetime (RUL) approaches for monitoring the health of concrete structures. Through the construction of a HI, the deterioration process can be processed and portrayed so that it can be forwarded to a prediction module for RUL prognosis. The degradation progression and failure can be identified by predicting the RUL based on the situation of the current specimen; as a result, maintenance can be planned to reduce safety risks, reduce financial costs, and prolong the specimen's useful lifetime. The portrayal of deterioration through HI construction from raw acoustic emission (AE) data is performed using a deep neural network (DNN), whose parameters are obtained by pretraining and fine tuning using a stack autoencoder (SAE). Kullback-Leibler divergence, which is calculated between a reference normal-conditioned signal and a current unknown signal, was used to represent the deterioration process of concrete structures, which has not been investigated for the concrete beams so far. The DNN-based constructor then learns to generate HI from raw data with KLD values as the training label. The HI construction result was evaluated with run-to-fail test data of concrete specimens with two measurements: fitness analysis of the construction result and RUL prognosis. The results confirm the reliability of KLD in portraying the deterioration process, showing a large improvement in comparison to other methods. In addition, this method requires no adept knowledge of the nature of the AE or the system fault, which is more favorable than model-based approaches where this level of expertise is compulsory. Furthermore, AE offers in-service monitoring, allowing the RUL prognosis task to be performed without disrupting the specimen's work.

This study proposes a fault diagnosis method (FD) for multistage centrifugal pumps (MCP) using informative ratio principal component analysis (Ir-PCA). To overcome the interference and background noise in the vibration signatures (VS) of the centrifugal pump, the fault diagnosis method selects the fault-specific frequency band (FSFB) in the first step. Statistical features in time, frequency, and wavelet domains were extracted from the fault-specific frequency band. In the second step, all of the extracted features were combined into a single feature vector called a multi-domain feature pool (MDFP). The multi-domain feature pool results in a larger dimension; furthermore, not all of the features are best for representing the centrifugal pump condition and can affect the condition classification accuracy of the classifier. To obtain discriminant features with low dimensions, this paper introduces a novel informative ratio principal component analysis in the third step. The technique first assesses the feature informativeness towards the fault by calculating the informative ratio between the feature within the class scatteredness and between-class distance. To obtain a discriminant set of features with reduced dimensions, principal component analysis was applied to the features with a high informative ratio. The combination of informative ratio-based feature assessment and principal component analysis forms the novel informative ratio principal component analysis. The new set of discriminant features obtained from the novel technique are then provided to the K-nearest neighbor (K-NN) condition classifier for multistage centrifugal pump condition classification. The proposed method outperformed existing state-of-the-art methods in terms of fault classification accuracy.

In this study, a scheme of remaining useful lifetime (RUL) prognosis from raw acoustic emission (AE) data is presented to predict the concrete structure's failure before its occurrence, thus possibly prolong its service life and minimizing the risk of accidental damage. The deterioration process is portrayed by the health indicator (HI), which is automatically constructed from raw AE data with a deep neural network pretrained and fine-tuned by a stacked autoencoder deep neural network (SAE-DNN). For the deep neural network structure to perform a more accurate construction of health indicator lines, a hit removal process with a one-class support vector machine (OC-SVM), which has not been investigated in previous studies, is proposed to extract only the hits which matter the most to the portrait of deterioration. The new set of hits is then harnessed as the training labels for the deep neural network. After the completion of the health indicator line construction, health indicators are forwarded to a long short-term memory recurrent neural network (LSTM-RNN) for the training and validation of the remaining useful life prediction, as this structure is capable of capturing the long-term dependencies, even with a limited set of data. Our prediction result shows a significant improvement in comparison with a similar scheme but without the hit removal process and other methods, such as the gated recurrent unit recurrent neural network (GRU-RNN) and the simple recurrent neural network.

This paper presents the design and development of a low cost and reliable maximal voluntary bite force sensor which can be manufactured in-house by using an acrylic laser cutting machine. The sensor has been designed for ease of fabrication, assembly, calibration, and safe use. The sensor is capable of use within an hour of commencing production, allowing for rapid prototyping/modifications and practical implementation. The measured data shows a good linear relationship between the applied force and the electrical resistance of the sensor. The output signal has low drift, excellent repeatability, and a large measurable range of 0 to 700 N. A high signal-to-noise response to human bite forces was observed, indicating the high potential of the proposed sensor for human bite force measurement.