Existing methods for tool wear recognition using online signal processing and deep learning techniques typically utilize support vector machines (SVM) and fully connected layers (FCL). These methods inherently struggle with capturing complex nonlinear wear patterns due to their dependence on linear transformations and fixed-weight architectures. To overcome these limitations, this study introduces a novel tool wear recognition method based on empirical wavelet transform (EWT) and Kolmogorov-Arnold Network (KAN). Through EWT, tool wear-related component signals are adaptively extracted from multi-sensor data such as cutting forces, accelerations, and acoustic emission signals. To further enhance feature extraction, this study designs a multi-scale convolutional network (MSCN) and an efficient channel attention (ECA) mechanism. The MSCN is aimed at isolating wear-sensitive features, while the ECA mechanism highlights critical wear indicators, thereby avoiding issues such as modal mixing and energy leakage. Based on these advancements, KAN has finally been adopted to construct the tool wear recognition model. Experimental results prove the feasibility of extracting tool wear component signals through EWT. Leveraging this foundation, the proposed method achieves superior recognition accuracy and feasibility compared to existing approaches. The average recognition accuracy of the proposed model exceeds 99.9%, confirming its effectiveness in tool wear recognition.

The study presents an innovative method for predicting stability lobe diagrams (SLDs) by explicitly incorporating the nonlinear dynamics of robotic milling systems. A frequency-domain decomposition (FDD)-based technique has been devised to systematically identify a comprehensive set of nonlinear frequency response functions (FRFs), from which a systematic mapping relationship between modal parameters, cutting forces, and feedrate is developed. Through this mapping, the complex governing equation of robotic milling is subsequently transformed into a more tractable form to predict SLDs. The significant advancement lies in that a decoupled formulation between modal parameters and excitation forces is established for the first time, thereby simplifying SLD analysis. It also offers the following two advantages over existing methods: (i) enhanced sensitivity in identifying the feedrate-dependent transition between regenerative chatter (RC) and low-frequency chatter (LFC), and (ii) substantial enhancement in the prediction accuracy of SLDs. Experimental results demonstrate that increasing feedrate shifts the RC-LFC transition point toward higher spindle speed regions, with more distinct variation patterns under strong-stiffness conditions. The proposed method achieves an average prediction accuracy of 89.48%, thereby effectively verifying its capability to enhance both the sensitivity and reliability of SLD predictions in robotic milling processes.

Chatter is a critical factor that limits machining quality and production efficiency in robotic milling. Therefore, timely and accurate online chatter detection is essential for achieving high-precision and high-efficiency machining. However, the dynamic properties of milling robots are highly pose-dependent, which significantly increases the randomness and complexity of chatter. Furthermore, due to the strong nonlinearity and flexibility of milling robots, chatter induces coupled three-dimensional vibrations. As a result, relying on vibration signals from a single direction is insufficient for achieving reliable chatter detection. In this paper, a multi-channel online chatter detection method in robotic milling is proposed, based on successive multivariate variational mode decomposition (SMVMD) and hybrid deep convolutional neural network. First, the SMVMD method combined with a dimensionless chatter index is employed to adaptively extract the chatter components in the three-channel acceleration signals. Then, the energy ratio sequence between the reconstructed chatter signal and the milling signal is calculated. Finally, a hybrid deep convolutional neural network (ICR-DCN), composed of Inception, convolutional block attention module (CBAM), residual network (ResNet) and classification module, is constructed to automatically extract chatter features from the energy ratio sequence and achieve online chatter detection. Experimental results demonstrate that the proposed method can achieve high-precision online chatter detection across variations in different robotic poses, cutting parameters, workpiece materials and tools. Comparative studies further validate the accuracy and stability of the proposed method across various scenarios.

Backlash in ball screw drives causes torque loss and engagement impact during motion reversal, resulting in tracking errors and oscillations. Existing compensation methods, which treat backlash as a lumped disturbance, fail to balance fast estimation and smooth compensation, leading to delayed or abrupt responses. This study introduces a two-layered composite control structure for backlash-affected ball screw drives, featuring a torque-loss compensation layer and an oscillation-damping layer to enhance tracking accuracy and suppress reversal oscillations. The backlash transition process is divided into three phases. Theoretical analysis demonstrates that tracking errors primarily occur in the first and third phases due to torque loss, while oscillations occur in the second phase due to engagement impact. A continuous prediction method for backlash transitions employs motor-side position increments after velocity reversal to achieve accurate phase switching and torque compensation. A combined backlash–friction feedforward strategy is developed to compensate for torque losses, while an extended state observer (ESO)-based controller rejects residual disturbances. An oscillation-damping layer with velocity-difference feedback suppresses engagement-induced vibrations, and a back-propagation artificial neural network characterizes position-dependent backlash using laser interferometer measurements. Comparative motion and milling tests validate the effectiveness of the proposed approach.

Hydroforming is an effective way for precision manufacturing of complex thin-walled components of aeroengines.According to the micro size characteristics of thin-walled superalloy C-shaped seal ring components of an aero-engine,a two-steps hydroforming process was proposed.The stress-strain analysis of multi-steps hydroforming process was conducted,and the finite element analysis model of multi-step hydroforming process was established.The impact of process variables,such as the height of the blank forming and the hydraulic loading path,on the forming quality of the seal ring was investigated using numerical simulation and process experimentation.Failure modes,such as the loss of section geometric characteristics,inadequate die attaching,and excessive wall thickness thinning,were also investigated.The process parameters were optimized.The results show that the two-step hydroforming process can achieve accurate forming of a thin-walled C-shaped metal seal ring.By using the optimized process parameters:height of blank forming of 1.0mm,first pass cavity pressure of 140 MPa,second pass cavity pressure of 180 MPa,the high-quality C-shaped seal ring with the degree of blank molding of 93.9%,thinning rate of 10.5%and wall thickness uniformity of 85.5%can be made.

Internally cooled milling tools deliver high-pressure cutting fluid directly to the tool-chip interface through internal channels and have been proven effective in mitigating the excessive heat generated when machining titanium alloys. However, conventional channel designs often suffer from poor surface smoothness and high flow resistance, which limit cooling efficiency. To address this, topology optimization combined with computational fluid dynamics (CFD) simulation is applied to minimize flow pressure drop and to improve the cooling performance of the milling tool in this study. CFD is employed to model the flow behavior of cutting fluid inside channels, and the volume of fluid (VOF) multiphase flow model is introduced to simulate the jetting process. The tool body is fabricated using selective laser melting (SLM) followed by mechanical post-processing. Milling tests are conducted on TC18 titanium alloy. CFD simulations and experimental results show that the optimized design reduces pressure drop within channels by 53.13%. Across various cutting parameters, cutting temperatures are reduced by 8.34% to 12.15%. Significant improvements are observed in surface roughness, tool wear, and chip morphology. The simulation and experimental trends are in good agreement, with notable improvements in outlet flow rate and velocity. These results validate topology-optimized internal channels as an effective strategy for enhancing milling tool cooling performance.

Existing chatter suppression methods using piezoelectric actuators face significant limitations in micro-milling, as they often require either complex multi-actuator setups, additional sensors for tool condition monitoring, or alterations to the spindle itself. To address these limitations, this study introduces an active chatter suppression method using a single piezoelectric actuator, which simplifies integration and enhances compatibility with space-constrained micro-milling conditions. By generating controlled feed-direction displacements that match the feed per tooth, the actuator disrupts the single-delay structure of micro-milling, introducing multiple time delays to expand the stable cutting zone. A critical challenge lies in the fact that the coupling between the rotating cutter and the workpiece, vibrating under actuator excitation, leads to uncertain initial contact positions and variable delay interactions. To ensure controllable suppression under these conditions, a sinusoidal modulation trajectory has been specifically designed. By precisely controlling the duration ratio and amplitudes of the positive/negative half-waves in each modulation cycle, the system ensures robust multiple-delay modulation, independent of the initial contact state. This ensures stable synchronization between the vibration trajectory of the workpiece and cutter tooth engagements. An open-loop system is proposed to achieve the preset workpiece vibration trajectory. The micro-milling dynamics is thereby formulated as a modulated multiple-delay chatter problem, with its stability subsequently analyzed through stability lobe diagrams (SLDs). To enhance computational efficiency, an optimization algorithm integrating a Bayesian approach with directional-factor pruning is presented for solving the SLDs. Several micro-milling tests demonstrate the effectiveness of the method.

The precise forming of complex thin-walled metallic components can be achieved through composite manufacturing process, where the macroscopic mechanical response and microstructural evolution exhibit significant coupling effects. A general multiscale sequential simulation framework was developed by coupling crystal plasticity finite element (CPFE) and cellular automaton (CA) models. A bidirectional grid mapping and data transfer method was established to address grid incompatibility and physical quantity mapping between different models. During the transfer from the CPFE model to the CA model, the proposed grid refinement mapping approach achieves lossless data transmission compared with the nearest-neighbor mapping method. In the reverse transfer from CA to CPFE, the average data transmission error is also nearly negligible when the coarsened element size approaches the CA cell size. The proposed multiscale simulation framework is applicable to both 2D and 3D conditions. For simulations of a two-stage uniaxial tension with intermediate annealing, the average prediction error of the 2D and 3D models is about 5%. Although the 3D model exhibits slightly improved prediction accuracy, the computational cost is approximately six times that of the 2D model. It indicates that the 2D model provides a reasonable balance between computational efficiency and predictive accuracy. Furthermore, the multiscale framework was applied to simulate the post-heat treatment process of additively manufactured alloy. The prediction errors for the recrystallized volume fraction and average grain size are both below 10%, and the stress-strain curves during subsequent uniaxial tension is predicted with an accuracy of approximately 95%. The results from the two application cases demonstrate that the proposed coupled model can accurately capture the mechanical response during deformation as well as the static recrystallization behavior during annealing, confirming the generality and reliability of the multiscale simulation framework.

Most existing control strategies for flexible ball screw feed drive systems (BSFDS) either rely on complex structures and extensive parameter adjustments to achieve vibration suppression and enhanced bandwidth, or maintain the conventional proportional-proportional-integral (P-PI) framework while grappling with significant parameter coupling between auxiliary compensators and the baseline controller. This study develops a structurally decoupled control approach for BSFDS operating within a P-PI full closed-loop architecture, aiming to suppress vibration and improve bandwidth performance by mathematically decoupling the velocity and position loop transfer functions. The analysis identifies that vibration mode and bandwidth limitations primarily stem from the flexibility in the velocity loop, while the flexibility-induced residual response in the position loop further restricts achievable bandwidth. Based on these findings, a shaping controller is integrated into the PI-controlled velocity loop. This controller employs structurally decoupled parameterization, with its parameters uniquely determined by the system's anti-resonance frequency and damping ratio. These settings are independent of the PI gains, effectively transforming the velocity loop into an equivalent second-order system with adjustable bandwidth characteristics. Building on this foundation, a position difference-based residual flexible-response compensation controller has been incorporated into the position loop to further extend the achievable bandwidth. Experimental results demonstrate the validity and effectiveness of the proposed approach.

Chatter is a type of self-excited vibration that occurs during machining, particularly when processing weak-rigidity aerospace parts, adversely affecting machining quality and efficiency. In order to prevent the detrimental effects of chatter in milling processes, a physically interpretable multi-order graph convolutional neural network (GCN)-based method is proposed for online chatter detection. In this method, the vertex and edge of the network are first constructed through the measured multi-channel vibration signal time series segments with short duration, establishing the groundwork for online chatter detection. Then, a multi-order GCN is developed by gathering feature information from multiple orders of neighbors of the target vertex to enhance its accuracy. Next, based on the multi-order GCN, a deep neural network is constructed and trained using the vibration data obtained by numerous milling experiments. During the training process, the edge weights are optimized by the gradient descent method. After that, an online chatter detection system is constructed, which demonstrates robust noise resistance through tests with different levels of noise interference added. Finally, experiments under various milling conditions validate that the proposed method can accurately detect milling chatter at its early weak stage, thereby successfully preventing its adverse effects.

This article presents a milling-based methodology for determining the fracture toughness, yield strength, and adhesion toughness of Inconel 718 under dynamic cutting conditions, based on an extension of Williams’ model. Unlike previous orthogonal cutting methods that require specialized setups, the proposed method employs a standard milling process, offering improved practicality and potential applicability in industrial environments. An analytical framework is developed by incorporating material fracture at the tool tip and adhesion at the tool-chip interface, in addition to plastic deformation and friction. The force-thickness relationship is extended to a three-dimensional oblique cutting configuration representative of milling. The normal shear angle is determined using a force minimization criterion, allowing the evaluation of cutting and transverse forces over a range of uncut chip thicknesses. Side milling experiments at different cutting speeds are conducted for validation. The measured cutting forces, combined with chip geometry, are used to extract the values of fracture toughness, yield strength, and adhesion toughness. The results exhibit consistent linear relationships, supporting the effectiveness of the proposed method. The main contribution of this work lies in the development of a practical and general methodology that extends force-based parameter identification to milling processes and enables the simultaneous evaluation of multiple mechanical parameters under realistic cutting conditions.

Existing milling residual stress models usually assume a static stress state following material removal and fail to account for the dynamic stress evolution that occurs on a tooth-by-tooth basis during milling. In fact, cutting actions associated with successive tooth cycles continuously disturb the stress equilibrium through coupled mechanisms of stress release, redistribution, and thermo-mechanical interactions, significantly impacting the final residual stress distribution. To address this limitation, this study proposes a model for predicting milling-induced residual stresses by effectively considering the dynamic redistribution occurring between individual cutter tooth cycles. Stress equilibrium relations at actual milling instants are established by comprehensively accounting for the initial workpiece stress state, as well as the coupling relationships among workpiece deformation, stresses in the removed material area, and newly generated stresses, based on moment balance conditions and a validated milling residual stress model. This model combines the time-varying stresses generated during the tooth-by-tooth material removal process with the initial residual stresses, thereby providing a precise depiction of how stresses evolve at different cutting instants under the combined influences of material removal and thermo-mechanical loading. Blind-hole method measurements, milling tests, and finite element simulations are performed on titanium alloy Ti6Al4V and aluminum alloy 7075 components, and good agreement between the predicted and measured residual stresses and their induced deformations confirm the correctness and reliability of the proposed model.

Tool runout alters the actual tool rotational center, which influences both the rotational radius and angular spacing of teeth relative to the spindle center, thereby changing the material removal process and significantly affecting both processing quality and tool lifespan. Traditional techniques for determining tool runout, regardless of whether they rely on static measurements or force analysis, often assume uniform pitch angles around the geometric axis and thus neglect the pitch-angle variations caused by actual rotational kinematics. This study presents a method for identifying tool runout parameters by considering the spindle-centered pitch distribution, utilizing measurable tooth-level responses, including actual tooth rotational kinematics, single tooth milling-based pitch measurements, and maximum rotational radii identified from slot milling tests. A kinematic model is developed to relate the theoretical pitch angle (with respect to the tool center) and the measured pitch angle (relative to the spindle center), along with rotational radii, through a geometric triangular configuration. Based on this model, analytical equations are derived linking the tool’s theoretical radius, maximum rotational radius, and runout offset, resulting in an explicit solution for identifying tool runout parameters. The maximum rotational radius is determined from slot widths measured during slot milling operations, while actual pitch angles are extracted by correlating tooth positions with their corresponding force signals in single-tooth milling tests. Experimental validations in both micro- and conventional-milling processes demonstrate that the identified tool runout parameters exhibit strong consistency with traditional methods. This validates the precision of the proposed approach and highlights its independence from specific cutting force magnitudes, ensuring robustness and applicability across various milling scenarios.

While existing one-step corner smoothing methods for five-axis G01 commands achieve greater computational efficiency compared to two-step approaches, they often employ jerk-limited profiles with only second-order continuity. This inherent jerk discontinuity adversely affects motion smoothness and surface quality. To overcome these limitations, this study presents a one-step formulation based on jerk-smooth kinematic profile blending, delivering superior motion smoothness. A trigonometric-based kinematic profile is developed as a smooth basis function, ensuring continuous differentiability of jerk while maintaining a concise analytical expression. To tackle the challenge of over-constrained kinematic parameters, a two-phase strategy is developed, involving geometry-driven prediction of the blending interval and decoupled adjustment using asymmetric profiles, enabling independent and systematic control of acceleration and deceleration phases. Furthermore, a geometric error control methodology is integrated to maximize smoothing errors within predefined limits. Using the tangential condition between resultant velocity and admissible area, an efficient method is constructed to determine the maximum blending time, which is applicable across various kinematic profiles. Experimental results, through comprehensive comparisons with representative approaches such as classical blending-based smoothing, advanced finite impulse response (FIR) filter-based smoothing, and high-order spline-based smoothing methods, demonstrate that the proposed method effectively improves motion smoothness and machining efficiency while strictly satisfying kinematic and geometric constraints.

Existing transmissibility-based operational modal analysis (TOMA) methods estimate in-process frequency response functions (FRFs) of milling spindle–tool systems directly from cutting responses, yet their accuracy is limited by the absence of explicit measurement noise modeling, resulting in biased estimates. This study introduces a parametric TOMA method tailored for identifying in-process FRFs in milling systems. The method explicitly addresses measurement noise by acquiring acceleration responses at multiple locations on non-rotating spindle components under actual cutting excitation. To counteract noise—modeled as white noise—a Frisch scheme integrated with high-order Yule–Walker (HOYW) equations is developed. By leveraging statistical independence between noise and the noise-free system response, the approach effectively isolates and suppresses noise while extracting true system dynamics. The transmissibility function is formulated via a polynomial matrix from a left matrix fraction description, enabling system pole identification through singular value decomposition and companion matrix techniques. Subsequently, in-process FRFs are reconstructed using the identified poles and residues obtained from impact testing. Milling tests demonstrate close agreement between identified and measured FRFs, enabling accurate stability lobe diagram (SLD) prediction and validating the method’s effectiveness and practicality.

Chatter is a critical limitation to productivity and machining quality in robotic milling. Accurate prediction of the pose-dependent tool point frequency response functions (FRFs) of milling robots is essential for effectively predicting and suppressing chatter. This paper presents a rapid prediction method for predicting the pose-dependent tool point dynamics of milling robots, incorporating cross receptances, which significantly influence both the dynamic behavior of milling robots and the stability of robotic milling processes. First, a comprehensive and generalized receptance coupling substructure analysis (RCSA) procedure is presented to couple the dynamics of the robot-spindle-holder-tool-shank (RSHTS) subsystem and cutting tools. Next, a surrogate model that combines proper orthogonal decomposition (POD) with multiple output Gaussian process regression (MOGPR) is developed to predict the pose-dependent receptances of the RSHTS subsystem. To facilitate accurate and efficient data collection, a measurement strategy using modal impact tests is introduced to acquire the full receptance matrix, including cross receptances. By preprocessing the measured receptance matrix with the POD method, the time-consuming step of extracting individual modal parameters is eliminated. Then the MOGPR model is used to exploit the inherent correlativity between different FRFs, significantly reducing the number of regression models compared to the Single Output Gaussian Process Regression (SOGPR) model while improving predictive performance and generalization capability. Finally, the presented method is validated through modal impact tests and milling tests conducted on an industrial robot. Experimental results confirm the accuracy, efficiency, and robustness of the presented method in predicting pose-dependent tool point dynamics. The effectiveness of the presented method is also demonstrated in enhancing stability predictions in robotic milling processes.

chatter detection models are established based on single classifiers and homogeneous ensemble classifiers, but the lack of diversity in these models leads to weak feature capturing and limited generalization capabilities. This article proposes a micro-milling chatter detection model based on stacking ensemble learning with diverse classifiers, aiming to enhance the model's generalization capability. The model captures data features from multiple perspectives to accurately classify machining states, including stable, slight, and severe chatter. The collection and processing of vibration signals are methodically established to obtain representative samples across different machining states. Base models are developed by training the samples through various classifiers: classification and regression tree (CART), k-nearest neighbor (KNN), feed-forward neural network (FNN), and gate recurrent unit (GRU). The final chatter detection model is constructed by integrating the outputs of these base models through a meta-classifier. The proposed model is rigorously validated through extensive micro-milling experiments, achieving a detection accuracy of (98.8 +/- 0.4)%, with a 95% confidence interval (CI) of [98.4%, 99.2%]. Its generalization capability is further evaluated under four different conditions: the Al-7050 workpiece but with different machining parameters, the Ti-6Al-4V workpiece, Tool 1, and Tool 2, demonstrating accuracies of (94.2 +/- 0.6)%, (90.3 +/- 1.2)%, (90.3 +/- 1.6)%, and (90.0 +/- 1.4)%, respectively, which significantly outperform those of support vector machine (SVM)-based models (e.g., traditional SVM and AdaBoost-SVM).

Pointer-type instruments are widely used in various industrial fields due to their simplicity, low cost, strong anti-interference capability, and durability. However, manually reading and recognizing these instruments can be complex and labor-intensive. To address this challenge, this paper proposes an automated method for pointer meter recognition and reading in complex environments, based on the WTOCA_ResNet framework, which integrates the WTFD, OCA, and ResNet modules. First, the improved WTOCA_ResNet model is employed to remove rain streaks from the input images. The instrument panel is then detected using YOLOv8. Tilt correction and dial region extraction are performed using a combination of SIFT feature matching, the RANSAC algorithm, perspective transformation, and Hough circle detection. To enhance dial details, bilateral filtering and CLAHE (Contrast Limited Adaptive Histogram Equalization) are applied. YOLOv8 is further utilized to locate the center of the dial, while the pointer tip is accurately identified using a combination of masking and Hough line detection. An improved multi-scale template matching algorithm is introduced to locate the zero scale, and the final reading is calculated based on angular measurements. Experimental results demonstrate that the WTOCA_ResNet model achieves state-of-the-art performance in image quality assessment after rain removal. The improved multi-scale template matching method reaches a recognition accuracy of 97.5%, and the overall error of the proposed method remains within 0.62% under real-world complex field conditions, indicating high accuracy, robustness, and practical applicability.

While many existing cutting force models achieve high predictive precision near the peak force, they frequently show substantial discrepancies in estimating the exiting moment during the tool's exiting stage. Despite its significance, this issue has garnered little attention and remains poorly understood. To address this knowledge gap, this study introduces a theoretical framework to explain these discrepancies, attributing them to the interplay of negative shearing effects, flank face interference, and workpiece deformation during the exiting stage of the cutting process. In the proposed model, the minimum energy principle is employed as the criterion for determining whether a negative or regular shearing effect occurs. A slip-line field is developed to model the negative shearing effect and its generated cutting forces. Flank interference, caused by the rapid elastic recovery of deflected cutters during the exiting stage of the cut, along with its induced interference force, is found and modeled using a combination of force equilibrium principles and constraints related to friction and acceleration limitations. The deformed workpiece, which is associated with burrs, is treated to follow the volume invariance principle and rotate around the intersection point of the instantaneous negative shearing plane and the workpiece boundary. The final cutting forces corresponding to different exiting instants together with the actual exiting moment are successfully determined by combining the effects of the negative shearing-induced component and flank interference-related component. The proposed model is validated through a series of cutting tests.

The existing body of research on multiple delays in micro milling has primarily focused analyzing the effects induced by tool runout. However, under varying machining parameters, the variations in static tool deflections caused by different cutting forces from tool runout will further influence the distribution range of multiple delay periods, thereby introducing additional complexities to the stability analysis of the machining process. This article proposes a dynamic model to characterize the chatter stability of micro milling with the consideration the coupled effects of cutter runout and tool deflections, emphasizing the advantage of taking account into multiple delays caused by both factors. Based on Timoshenko beam theory and radial runout model, an iterative algorithm is constructed to calculate the tool's actual effective radii under deflections. Subsequently, instantaneous directional factors involved in the dynamic model are reconstructed with the aid of the obtained multiple delays and actual tool's radii. Finally, a numerical algorithm aiming at efficiently solving the stability lobe diagrams (SLDs) constructed based on Newton-Raphson method, greatly reducing the computation time required by the traditional semi-discretization method. A series of micro milling experiments verify proposed model.