Modern rotating machines, which are represented by high-end grinding machines, have developed toward high precision, intelligence, and high durability in recent years. As the core components of grinding machine spindles, grinding wheels greatly affect the vibration level during operation. The unbalance vibration self-recovery regulation (UVSRR) system is proposed to suppress the vibration of grinding wheels during workpiece processing, eliminating or minimizing the imbalance. First, technical principles and the system composition are introduced. Second, the balancing actuator in the UVSRR system is analyzed in detail. The advanced nature of the improved structure is presented through structure introduction and advantage analysis. The performance of the balancing actuator is mutually verified by the theoretical calculation of torque and software simulation. Results show that the self-locking torque satisfies the actual demand, and the driving torque is increased by 1.73 times compared with the traditional structure. Finally, the engineering application value of the UVSRR system is verified by laboratory performance comparison and actual factory application. The balancing speed and effect of the UVSRR system are better than those of an international mainstream product and, the quality of the workpieces machined in the factory improved by 40%.

Impact force identification is important for structure health monitoring especially in applications involving composite structures. Different from the traditional direct measurement method, the impact force identification technique is more cost effective and feasible because it only requires a few sensors to capture the system response and infer the information about the applied forces. This technique enables the acquisition of impact locations and time histories of forces, aiding in the rapid assessment of potentially damaged areas and the extent of the damage. As a typical inverse problem, impact force reconstruction and localization is a challenging task, which has led to the development of numerous methods aimed at obtaining stable solutions. The classical {\ell }_2 regularization method often struggles to generate sparse solutions. When solving the under-determined problem, {\ell }_2 regularization often identifies false forces in non-loaded regions, interfering with the accurate identification of the true impact locations. The popular {\ell }_1 sparse regularization, while promoting sparsity, underestimates the amplitude of impact forces, resulting in biased estimations. To alleviate such limitations, a novel non-convex sparse regularization method that uses the non-convex {\ell }_{1-2} penalty, which is the difference of the {\ell }_1 and {\ell }_2 norms, as a regularizer, is proposed in this paper. The principle of alternating direction method of multipliers (ADMM) is introduced to tackle the non-convex model by facilitating the decomposition of the complex original problem into easily solvable subproblems. The proposed method named {\ell }_{1-2}\text{-ADMM} is applied to solve the impact force identification problem with unknown force locations, which can realize simultaneous impact localization and time history reconstruction with an under-determined, sparse sensor configuration. Simulations and experiments are performed on a composite plate to verify the identification accuracy and robustness with respect to the noise of the {\ell }_{1-2}\text{-ADMM} method. Results indicate that compared with other existing regularization methods, the {\ell }_{1-2}\text{-ADMM} method can simultaneously reconstruct and localize impact forces more accurately, facilitating sparser solutions, and yielding more accurate results.

One of the core challenges of intelligent fault diagnosis is that the diagnosis model requires numerous labeled training datasets to achieve satisfactory performance. Generating training data using a virtual model is a potential solution for addressing such a problem, and the construction of a high-fidelity virtual model is fundamental and critical for data generation. In this study, a digital twin-assisted dynamic model updating method for fault diagnosis is thus proposed to improve the fidelity and reliability of a virtual model, which can enhance the generated data quality. First, a virtual model is established to mirror the vibration response of a physical entity using a dynamic modeling method. Second, the modeling method is validated through a frequency analysis of the generated signal. Then, based on the signal similarity indicator, a physical–virtual signal interaction method is proposed to dynamically update the virtual model in which parameter sensitivity analysis, surrogate technique, and optimization algorithm are applied to increase the efficiency during the model updating. Finally, the proposed method is successfully applied to the dynamic model updating of a single-stage helical gearbox; the virtual data generated by this model can be used for gear fault diagnosis.

Recently, advanced sensing techniques ensure a large number of multivariate sensing data for intelligent fault diagnosis of machines. Given the advantage of obtaining accurate diagnosis results, multi-sensor fusion has long been studied in the fault diagnosis field. However, existing studies suffer from two weaknesses. First, the relations of multiple sensors are either neglected or calculated only to improve the diagnostic accuracy of fault types. Second, the localization for multi-source faults is seldom investigated, although locating the anomaly variable over multivariate sensing data for certain types of faults is desirable. This article attempts to overcome the above weaknesses by proposing a global method to recognize fault types and localize fault sources with the help of multi-sensor relations (MSRs). First, an MSR model is developed to learn MSRs automatically and further obtain fault recognition results. Second, centrality measures are employed to analyze the MSR graphs learned by the MSR model, and fault sources are therefore determined. The proposed method is demonstrated by experiments on an induction motor and a centrifugal pump. Results show the proposed method’s validity in diagnosing fault types and sources.

As parameter independent yet simple techniques, the energy operator (EO) and its variants have received considerable attention in the field of bearing fault feature detection. However, the performances of these improved EO techniques are subjected to the limited number of EOs, and they cannot reflect the non-linearity of the machinery dynamic systems and affect the noise reduction. As a result, the fault-related transients strengthened by these improved EO techniques are still subject to contamination of strong noises. To address these issues, this paper presents a novel EO fusion strategy for enhancing the bearing fault feature nonlinearly and effectively. Specifically, the proposed strategy is conducted through the following three steps. First, a multi-dimensional information matrix (MDIM) is constructed by performing the higher order energy operator (HOEO) on the analysis signal iteratively. MDIM is regarded as the fusion source of the proposed strategy with the properties of improving the signal-to-interference ratio and suppressing the noise in the low-frequency region. Second, an enhanced manifold learning algorithm is performed on the normalized MDIM to extract the intrinsic manifolds correlated with the fault-related impulses. Third, the intrinsic manifolds are weighted to recover the fault-related transients. Simulation studies and experimental verifications confirm that the proposed strategy is more effective for enhancing the bearing fault feature than the existing methods, including HOEOs, the weighting HOEO fusion, the fast Kurtogram, and the empirical mode decomposition.

Blade strain distribution and its change with time are crucial for reliability analysis and residual life evaluation in blade vibration tests. Traditional strain measurements are achieved by strain gauges (SGs) in a contact manner at discrete positions on the blades. This study proposes a method of full-field and real-time strain reconstruction of an aero-engine blade based on limited displacement responses. Limited optical measured displacement responses are utilized to reconstruct the full-field strain. The full-field strain distribution is in-time visualized. A displacement-to-strain transformation matrix is derived on the basis of the blade mode shapes in the modal coordinate. The proposed method is validated on an aero-engine blade in numerical and experimental cases. Three discrete vibrational displacement responses measured by laser triangulation sensors are used to reconstruct the full-field strain over the whole operating time. The reconstructed strain responses are compared with the results measured by SGs and numerical simulation. The high consistency between the reconstructed and measured results demonstrates the accurate strain reconstructed by the method. This paper provides a low-cost, real-time, and visualized measurement of blade full-field dynamic strain using displacement response, where the traditional SGs would fail.

As a nondestructive testing technique, terahertz time-domain spectroscopy technology is commonly used to measure the thickness of ceramic coat in thermal barrier coatings (TBCs). However, the invisibility of ceramic/thermally grown oxide (TGO) reflective wave leads to the measurement failure of natural growth TGO whose thickness is below 10 μm in TBCs. To detect and monitor TGO in the emergence stage, a time of flight (TOF) improved TGO thickness measurement method is proposed. A simulative investigation on propagation characteristics of terahertz shows the linear relationship between TGO thickness and phase shift of feature wave. The accurate TOF increment could be acquired from wavelet soft threshold and cross-correlation function with negative effect reduction of environmental noise and system oscillation. Thus, the TGO thickness could be obtained efficiently from the TOF increment of the monitor area with different heating times. The averaged error of 1.61 μm in experimental results demonstrates the highly accurate and robust measurement of the proposed method, making it attractive for condition monitoring and life prediction of TBCs.

Physical models carry quantitative and explainable expert knowledge. However, they have not been introduced into gas face seal diagnosis tasks because of the unacceptable computational cost of inferring the input fault parameters for the observed output or solving the inverse problem of the physical model. The presented work develops a surrogate-model-assisted method for solving the nonlinear inverse problem in limited physical model evaluations. The method prepares a small initial database on sites generated with a Latin hypercube design and then performs an iterative routine that benefits from the rapidity of the surrogate models and the reliability of the physical model. The method is validated on simulated and experimental cases. Results demonstrate that the method can effectively identify the parameters that induce the abnormal signal output with limited physical model evaluations. The presented work provides a quantitative, explainable, and feasible approach for identifying the cause of gas face seal contact. It is also applicable to mechanical devices that face similar difficulties.

Axial piston pumps have wide applications in hydraulic systems for power transmission. Their condition monitoring and fault diagnosis are essential in ensuring the safety and reliability of the entire hydraulic system. Vibration and discharge pressure signals are two common signals used for the fault diagnosis of axial piston pumps because of their sensitivity to pump health conditions. However, most of the previous fault diagnosis methods only used vibration or pressure signal, and literatures related to multi-sensor data fusion for the pump fault diagnosis are limited. This paper presents an end-to-end multi-sensor data fusion method for the fault diagnosis of axial piston pumps. The vibration and pressure signals under different pump health conditions are fused into RGB images and then recognized by a convolutional neural network. Experiments were performed on an axial piston pump to confirm the effectiveness of the proposed method. Results show that the proposed multi-sensor data fusion method greatly improves the fault diagnosis of axial piston pumps in terms of accuracy and robustness and has better diagnostic performance than other existing diagnosis methods.

Convolutional neural network (CNN) has achieved remarkable applications in fault diagnosis. However, the tuning aiming at obtaining the well-trained CNN model is mainly manual search. Tuning requires considerable experiences on the knowledge on CNN training and fault diagnosis, and is always time consuming and labor intensive, making the automatic hyper parameter optimization (HPO) of CNN models essential. To solve this problem, this paper proposes a novel automatic CNN (ACNN) for fault diagnosis, which can automatically tune its three key hyper parameters, namely, learning rate, batch size, and L2-regulation. First, a new deep reinforcement learning (DRL) is developed, and it constructs an agent aiming at controlling these three hyper parameters along with the training of CNN models online. Second, a new structure of DRL is designed by combining deep deterministic policy gradient and long short-term memory, which takes the training loss of CNN models as its input and can output the adjustment on these three hyper parameters. Third, a new training method for ACNN is designed to enhance its stability. Two famous bearing datasets are selected to evaluate the performance of ACNN. It is compared with four commonly used HPO methods, namely, random search, Bayesian optimization, tree Parzen estimator, and sequential model-based algorithm configuration. ACNN is also compared with other published machine learning (ML) and deep learning (DL) methods. The results show that ACNN outperforms these HPO and ML/DL methods, validating its potential in fault diagnosis.

Gear wear is one of the most common gear failures, which changes the mesh relationship of normal gear. A new mesh relationship caused by gear wear affects meshing excitations, such as mesh stiffness and transmission error, and further increases vibration and noise level. This paper aims to establish the model of mesh relationship and reveal the vibration characteristics of external spur gears with gear wear. A geometric model for a new mesh relationship with gear wear is proposed, which is utilized to evaluate the influence of gear wear on mesh stiffness and unloaded static transmission error (USTE). Based on the mesh stiffness and USTE considering gear wear, a gear dynamic model is established, and the vibration characteristics of gear wear are numerically studied. Comparison with the experimental results verifies the proposed dynamic model based on the new mesh relationship. The numerical and experimental results indicate that gear wear does not change the structure of the spectrum, but it alters the amplitude of the meshing frequencies and their sidebands. Several condition indicators, such as root-mean-square, kurtosis, and first-order meshing frequency amplitude, can be regarded as important bases for judging gear wear state.

The fault diagnosis of bearings is crucial in ensuring the reliability of rotating machinery. Deep neural networks have provided unprecedented opportunities to condition monitoring from a new perspective due to the powerful ability in learning fault-related knowledge. However, the inexplicability and low generalization ability of fault diagnosis models still bar them from the application. To address this issue, this paper explores a decision-tree-structured neural network, that is, the deep convolutional tree-inspired network (DCTN), for the hierarchical fault diagnosis of bearings. The proposed model effectively integrates the advantages of convolutional neural network (CNN) and decision tree methods by rebuilding the output decision layer of CNN according to the hierarchical structural characteristics of the decision tree, which is by no means a simple combination of the two models. The proposed DCTN model has unique advantages in 1) the hierarchical structure that can support more accuracy and comprehensive fault diagnosis, 2) the better interpretability of the model output with hierarchical decision making, and 3) more powerful generalization capabilities for the samples across fault severities. The multiclass fault diagnosis case and cross-severity fault diagnosis case are executed on a multicondition aeronautical bearing test rig. Experimental results can fully demonstrate the feasibility and superiority of the proposed method.

Existing fault diagnosis methods usually assume that there are balanced training data for every machine health state. However, the collection of fault signals is very difficult and expensive, resulting in the problem of imbalanced training dataset. It will degrade the performance of fault diagnosis methods significantly. To address this problem, an imbalanced fault diagnosis of rotating machinery using autoencoder-based SuperGraph feature learning is proposed in this paper. Unsupervised autoencoder is firstly used to compress every monitoring signal into a low-dimensional vector as the node attribute in the SuperGraph. And the edge connections in the graph depend on the relationship between signals. On the basis, graph convolution is performed on the constructed SuperGraph to achieve imbalanced training dataset fault diagnosis for rotating machinery. Comprehensive experiments are conducted on a benchmarking publicized dataset and a practical experimental platform, and the results show that the proposed method can effectively achieve rotating machinery fault diagnosis towards imbalanced training dataset through graph feature learning.

Deep learning has achieved much success in mechanical intelligent fault diagnosis in recent years. However, many deep learning methods cannot fully extract fault information to recognize mechanical health states when processing high-dimensional samples. Therefore, a multi-model ensemble deep learning method based on deep convolutional neural network (DCNN) is proposed in this study to accomplish fault recognition of high-dimensional samples. First, several 1D DCNN models with different activation functions are trained through dimension reduction learning to obtain different fault features from high-dimensional samples. Second, the obtained features are constructed into 2D images with multiple channels through a conversion method. The integrated 2D feature images can effectively represent the fault characteristic contained in raw high-dimension vibration signals. Lastly, a 2D DCNN model with multi-layer convolution and pooling is used to automatically learn features from the 2D images and identify the fault mode of the mechanical equipment by adopting a softmax classifier. The proposed method, which is validated using the bearing public dataset of Case Western Reserve University, USA and a one-stage reduction gearbox dataset, has high recognition accuracy. Compared with other classical deep learning methods, the proposed fault diagnosis method has considerable improvements.