Frequency invariant classification of ultrasonic weld inspection signals

Machine learning for ultrasonic nondestructive examination of welding defects: A systematic review

Recent years have seen a substantial increase in the application of machine learning (ML) for automated analysis of nondestructive examination (NDE) data. One of the applications of interest is the use of ML for the analysis of data from in-service inspection of welds in nuclear power and other industries. These types of inspections are performed in accordance with criteria described in the ASME Boiler and Pressure Vessel Code and require the use of reliable NDE techniques. The rapid growth in ML methods and the diversity of possible approaches indicate a need to assess the current capabilities of ML and automated data analysis for NDE and identify any gaps or shortcomings in current ML technologies as applied to the automated analysis of NDE data. In particular, there is a need to determine the impact of ML on the NDE reliability. This paper discusses the findings from a literature survey on the current state of ML for the automated analysis of data from ultrasonic NDE of weld flaws. It discusses an overview of ultrasonic NDE as used for weld inspections in nuclear power and other industries. Data sets and ML models used in the literature are summarized, along with a generally applicable workflow for ML. Findings on the capabilities, limitations and potential gaps in feature selection, data selection, and ML model optimization are discussed. The paper identified several needs for quantifying and validating the performance of ML methods for ultrasonic NDE, including the need for common data sets.

Improved Ultrasonic Dead Zone Detectability of Work Rolls Using a Convolutional Neural Network

Rolled carbon steel sheets used in industrial and construction sites are formed by passing metal stock between two rotating rolls using a rolling mill, and the work roll is an essential part of the rolling mill. As the work rolls are in direct contact with the workpiece, the process quality is highly sensitive to their surface integrity, which is maintained through rough and finish cuttings; ultrasonic inspection is often performed after rough cutting the surfaces of work rolls. Ultrasonic inspection signals comprise signals reflected from and below the surface. Depending on the size of the subsurface defects, the thickness of the finish cutting is determined. The signals reflected by defects close to the surface overlap with those from the work roll surface, which is referred to as the ultrasonic dead zone and makes defect detection difficult. Since visual detection of flaws is not possible from signals collected from the dead zone, finish cutting is commonly performed up to the dead zone depth; this requires unnecessary cost and process time, which must be improved. Therefore, in this study a convolutional neural network is used to improve defect detection performance in the ultrasonic dead zone during the inspection of work rolls.

System Invariant Method for Ultrasonic Flaw Classification in Weldments Using Residual Neural Network

The industrial use of ultrasonic flaw classification using neural networks in weldments must overcome many challenges. A major constraint is the use of numerous systems, including a combination of transducers and equipment. This causes high complexity in the datasets used in the training of neural networks, which decreases performance. In this study, the performance of a neural network was enhanced using signal processing on an ultrasonic weldment flaw dataset to achieve system invariance. The dataset contained 5,839 ultrasonic flaw signals collected by various types of transducers connected to KrautKramer USN60. Every signal in the dataset was from 45 FlawTech/Sonaspection weldment specimens with five types of flaw: crack, lack of fusion, slag inclusion, porosity, and incomplete penetration. The neural network used in this study is a residual neural network with 19 layers. The performance evaluation of the same network structure showed that the original database can achieve 62.17% ± 4.13% accuracy, and that the invariant database using the system invariant method can achieve 91.45% ± 1.77% accuracy. The results demonstrate that using a system invariant method for ultrasonic flaw classification in weldments can improve the performance of a neural network with a highly complex dataset.

Defect Identification of Pipeline Ultrasonic Inspection Based on Multi-Feature Fusion and Multi-Criteria Feature Evaluation

This paper presents a novel model for ultrasonic defect identification relying on multi-feature fusion and multi-criteria feature evaluation (MFF-MCFE). Based on feature extraction, feature selection, pattern recognition and data fusion algorithm, this model analyzes ultrasonic echo signal data from single-probe ultrasonic inspection, and based on wavelet packet transform (WPT), empirical mode decomposition (EMD) and discrete wavelet transform (DWT), the main features from the collected ultrasonic echo signals are also extracted. These features are also evaluated by means of Representation Entropy (RE), Fisher’s ratio (FR) and Mahalanobis distance (MD), and the results are fused with Dempster–Shafer (D-S) evidence theory and the corresponding feature subsets are formed according to the fusion result. The support vector machine (SVM) is used as the classifier to recognize the defect signal, and the subsequent classification results are integrated by D-S evidence theory, which leads to the final recognition results. On this basis, a series of experiments were carried out to compare the performance of the developed model with that of the models using single feature sets and single feature evaluation criterion. Meanwhile, the principal component analysis (PCA) was also involved in the corresponding comparative analysis. The experimental results showed that this model is suitable for the identification and diagnosis of pipeline defects, and its classification accuracy could be reached up to 96.29% with stronger robustness and stability.

A Novel Electrocardiogram Biometric Identification Method Based on Temporal-Frequency Autoencoding

For good performance, most existing electrocardiogram (ECG) identification methods still need to adopt a denoising process to remove noise interference beforehand. This specific signal preprocessing technique requires great efforts for algorithm engineering and is usually complicated and time-consuming. To more conveniently remove the influence of noise interference and realize accurate identification, a novel temporal-frequency autoencoding based method is proposed. In particular, the raw data is firstly transformed into the wavelet domain, where multi-level time-frequency representation is achieved. Then, a prior knowledge-based feature selection is proposed and applied to the transformed data to discard noise components and retain identity-related information simultaneously. Afterward, the stacked sparse autoencoder is introduced to learn intrinsic discriminative features from the selected data, and Softmax classifier is used to perform the identification task. The effectiveness of the proposed method is evaluated on two public databases, namely, ECG-ID and Massachusetts Institute of Technology-Biotechnology arrhythmia (MIT-BIH-AHA) databases. Experimental results show that our method can achieve high multiple-heartbeat identification accuracies of 98.87%, 92.3%, and 96.82% on raw ECG signals which are from the ECG-ID (Two-recording), ECG-ID (All-recording), and MIT-BIH-AHA database, respectively, indicating that our method can provide an efficient way for ECG biometric identification.

Non-destructive evaluation of depth of surface cracks using ultrasonic frequency analysis

Ultrasonic is one of the most common uses of a non-destructive evaluation method for crack detection and characterization. The effectiveness of the acoustic-ultrasound Structural Health Monitoring (SHM) technique for the determination of the depth of the surface crack was presented. A method for ultrasonic sizing of surface cracks combined with the time domain and frequency spectrum was adopted. The ultrasonic frequency spectrum was obtained by Fourier transform technique. A series of test specimens with various depths of surface crack ranging from 1 mm to 8 mm was fabricated. The depth of the surface crack was evaluated using the pulse-echo technique. In this work, three different longitudinal waves with frequencies of 2.25 MHz, 5 MHz and 10 MHz were employed to investigate the effect of frequency on the sizing detection of surface cracks. Reasonable accuracies were achieved with measurement errors less than 7%.

Autonomous corrosion detection in gas pipelines: a hybrid-fuzzy classifier approach using ultrasonic nondestructive evaluation protocols

In this paper, a customized classifier is presented for the industry-practiced nondestructive evaluation (NDE) protocols using a hybrid-fuzzy inference system (FIS) to classify the corrosion and distinguish it from the geometric defects or normal/healthy state of the steel pipes used in the gas/petroleum industry. The presented system is hybrid in the sense that it utilizes both soft computing through fuzzy set theory, as well as conventional parametric modeling through H(infinity) optimization methods. Due to significant uncertainty in the power spectral density of the noise in ultrasonic NDE procedures, the use of optimal H(2) estimators for defect characterization is not so accurate. A more appropriate criterion is the H(infinity) norm of the estimation error spectrum which is based on minimization of the magnitude of this spectrum and hence produces more robust estimates. A hybrid feature set is developed in this work that corresponds to a) geometric features extracted directly from the raw ultrasonic A-scan data (which are the ultrasonic echo pulses in 1-Dtraveling inside the metal perpendicular to its 2 surfaces) and b) mapped features from the impulse response of the estimated model of the defect waveform under study. An experimental strategy is first outlined, through which the necessary data are collected as A-scans. Then, using the H(infinity) estimation approach, a parametric transfer function is obtained for each pulse. In this respect, each A-scan is treated as output from a defining function when a pure/healthy metal's A-scan is used as its input. Three defining states are considered in the paper; healthy, corroded, and defective, where the defective class represents metal with artificial or other defects. The necessary features are then calculated and are then supplied to the fuzzy inference system as input to be used in the classification. The resulting system has shown excellent corrosion classification with very low misclassification and false alarm rates.

Lamb waves propagation in elastic plane layers with a joint strip

The Lamb waves are used for the ultrasonic characterization of welds because of their relative long-range propagation. In this paper, a simplified model of a weld-strip between two identical semi-infinite elastic layers is investigated. The reflected and transmitted ultrasonic fields are expressed by modal series whose coefficients are obtained by application of orthogonality relation. Comparisons with solutions obtained by finite elements wave propagation simulations are made. The energy balance between the incident and the scattered waves is also used to verify the accuracy of the obtained modal amplitudes.

An incremental learning algorithm with confidence estimation for automated identification of NDE signals

An incremental learning algorithm is introduced for learning new information from additional data that may later become available, after a classifier has already been trained using a previously available database. The proposed algorithm is capable of incrementally learning new information without forgetting previously acquired knowledge and without requiring access to the original database, even when new data include examples of previously unseen classes. Scenarios requiring such a learning algorithm are encountered often in nondestructive evaluation (NDE) in which large volumes of data are collected in batches over a period of time, and new defect types may become available in subsequent databases. The algorithm, named Learn++, takes advantage of synergistic generalization performance of an ensemble of classifiers in which each classifier is trained with a strategically chosen subset of the training databases that subsequently become available. The ensemble of classifiers then is combined through a weighted majority voting procedure. Learn++ is independent of the specific classifier(s) comprising the ensemble, and hence may be used with any supervised learning algorithm. The voting procedure also allows Learn++ to estimate the confidence in its own decision. We present the algorithm and its promising results on two separate ultrasonic weld inspection applications.