Single-crystal silicon is widely used as a substrate material for integrated circuits. Typically, the wafer is cut by a wire saw with fixed diamond abrasive owning to the wire saw’s narrow kerf and low cutting force. However, the wafer has a large number of scratches by contact cutting with the wire saw, and the material removal rate (MRR) of the wire saw still needs to be improved. This paper proposes the abrasion-assisted wire electrical discharge machining (A-WEDM) process for single-crystal silicon and investigates the micro-cutting mechanism of A-WEDM by collecting and analyzing the discharge waveform during processing, observing the workpiece surface topography after processing, and comparing the diamond protrusion height and the discharge gap width. In addition, the paper also compares the difference in process performance between the wire saw machining and A-WEDM. The experimental results show that the maximum discharge gap is 57 μm, i.e., the distance between the workpiece and wire core where discharge first occurs in the initial stage of machining. When the machining is in a stable stage, within each voltage pulse material is removed under the coupling effect of discharge and diamond abrasive grinding. In each pulse interval, the electrical discharge effect stops, but the diamond abrasive grinding action still exists, thereby removing part of the recast layer and craters generated by the electrical discharge. Compared with the wire saw machining, A-WEDM has fewer scratches and higher MRR, however, the surface roughness is higher than when using wire saw machining alone.

Camera market has been rapidly expanding in industrial use such as broadcasting, on-vehicle, security and medical one, while simultaneously and strongly requiring enhancement of the image quality. To meet the requirement, glass lenses as key devices in cameras must be fabricated with higher degree of form accuracy of surfaces. Surface of glass lenses is formed and finished by polishing using small tools. The existing polishing technologies, however, have serious problems including an unstable material removal rate with the accumulated polishing time. In this study, we conducted glass-polishing experiments to investigate the variation mechanism of the material removal rate. Based on the findings, a new polishing pad was proposed. Our experiments confirmed that the proposed polishing pad was effective in suppressing the variation in the material removal rate.

Laser metal-wire deposition (LMwD) is suitable for fabricating large-scale components and near-net shape with high deposition rates and low costs. However, uncontrollable mechanical properties of as-built samples are crucial problems that need to be investigated. In the present study, by discussing the microstructure of joint area between layers of Ti6Al4V thin wall as-built samples, the deposition regularity of LMwD is studied. The results show that the upper layer can re-melt the top side of the lower area and change its initial texture. The distribution of heat affected zone (HAZ) between layers is elucidated clearly. The provided data makes a stable and better microstructure and mechanical properties become possible.

In this paper, changes on electroplated diamond wheel were investigated during the trimming process of CFRP. The influences of the wheel speed to the wheel surface are examined. Fine ground surfaces and small dimensional errors proved the high machinability of the electroplated wheels. But the experimental results also showed that low speed of wheel causes excessive chip loading for very short process time and increases the grinding forces rapidly.

Recently, many wearing simulation systems, which do not need samples of clothes or actual trying on them, have been developed to improve the productivity of clothes. However, in case of knitted clothes, conventional systems only offer the looking not based on mechanical consideration at the stitch level because such consideration leads to a significant increase of computational cost. In this paper, we propose a shape prediction method of knitted stitched using machine learning. First, a yarn is modeled as a structure with straight springs, rotating springs, and torsion springs. By minimizing the potential energy of yarns, the stable shape of a stitch can be derived. Next, using such shapes as training data, machine learning was performed with nonlinear neural networks. Then, various shapes of the stitch can be predicted without time-consuming optimization. Our proposed method will be useful for precise wearing simulation of knitted clothes.

As product systems become larger and more complex, their design space is increasing, and it is becoming more difficult to optimize them. In the past, hierarchical optimization methods have been proposed to solve this problem by decomposing the entire system into several subsystems and dividing and reducing the design space. However, they are ineffective for problems with strong dependencies among many subsystems. On the other hand, it has become clear that reinforcement learning can be used to efficiently search for optimal solutions to optimization problems such as solving small-scale games. If we try to apply reinforcement learning to large-scale optimization problems, reinforcement learning space increases, and optimization becomes difficult. Therefore, in this study, we used multi-agent reinforcement learning to reduce the learning space handled by reinforcement learning at a time. In this method, we propose an algorithm that reduces the learning space by allowing each agent to learn only the priority of a subsystem based on the relationship between the design variables and the evaluation values of each subsystem. Finally, the proposed method reduced the number of evaluations to less than 10% of the conventional number while maintaining the solution quality of the optimization.

Conventional mechanical machining processes such as milling, turning, and grinding are position-dependent processes where the volume of material removed is solely dependent on the engagement geometry of the tool and workpiece. However, advanced precision polishing processes with a compliant tool, fluid jet, or laser are characterized as time-dependent processes (TDP) where the volume of material removed depends not only on the tool position but also on its influence duration on the workpiece. Accordingly, accurate planning of the feedrate profile in sync with the tool position becomes necessary to accurately control the material removal rate and final target shape profile. A novel feedrate planning algorithm is proposed in this paper, which schedules the position-dependent feedrate based on the time-dependent process dynamics and the target shape profile. The feedrate profile is designed as a B-spline, and control points are optimized in the sense of least squares based on the process dynamics and target final shape. The developed algorithms were demonstrated in fluid jet polishing of precision optics.

A method to predict a performance of a developable surface using Gaussian Process Regression is proposed. A developable surface, which can be developed without stretching, are widely used in many products, such as shipbuilding or architecture. The performance of such products, such as evaluated by finite element analysis, is not always formulated using their geometry, which designing the shape as designers intend depends on their skills. Therefore, it is necessary to support such design process. For efficiency of its design process, this paper proposes a method to predict the performance of a developable surface by applying machine learning. Upon applying, features extracted from its shape based on the differential geometry and gaussian process regression is used. To verify the validity of our proposed method, we applied it to the analysis data of FEM in the design process of a brassiere cup and considered the regression accuracy.

In this paper, a weighted item collecting problem in a directed bipartite graph is considered, where one finite set consists of items with profits and the other consists of players with costs. A budget constraint is imposed on the players with respect to the costs paid by them in an arc reversing strategy. The objective is to find a feasible arc reversing strategy which maximizes the total profit of collected items. The problem can be viewed as a generalization of a constrained via minimization in integrated circuit design automation. In this paper, a faster implementation of a heuristic algorithm based on simulated annealing is investigated. For this, some mathematical observations are first discussed. Then, the observations are utilized in an existing simulated annealing heuristic so as to make it run faster. Numerical experiments are conducted to demonstrate the presented implementation, and the results are reported.

In recent production systems, it is required to increase the profits while responding to various user's preferences and short due date. To deal with these situations, it is important to increase the factory productivity with production cost reduction. Our group focuses on the factory load adjustment by using outsourced factories and adjusting the number of usage machines in the factory in order to improve the productivity and reduce the production cost. In previous our work, the production scheduling method using a linear weighted sum technique was proposed for minimizing the tardiness and reducing the production cost. However, in the previous method, it was difficult to adjust the appropriate weighted parameters and to evaluate the tardiness and production cost directly. In this proceeding, the multi-objective production scheduling method using a simulated annealing technique is proposed for solving the previous issues and presenting the various possibilities of factory load adjustment. In the computational experiments for flexible job shop scheduling problem, the ability of proposed method is evaluated.

When making a mold, a high-quality machined surface or good appearance is required. However, there is no established method for evaluating the quality of a machined surface, and such judgements are generally based on subjective evaluations. In this study, the relation between frequency analysis and surface quality was investigated by comparing the results of frequency analysis of profile curves changing the tool radius and pick feed of a ball end mill with the sensory scale of surface quality measured by the Thurstone pairwise comparison method. The results showed that the effect of frequency was stronger than the effect of the main amplitude spectrum on the machined surface quality under the experimental conditions.

In recent years, sensorless cutting force estimation technology for process monitoring in real time has been attracting attention against the background of Industry 4.0. Sensorless cutting force estimation by a disturbance observer is effective for monitoring machining conditions. However, extension to a dual-inertial system is a practical limit. Therefore, in this research, a Kalman-filter based cutting force estimation methodology from the encoder information is proposed applying the modal parameters identified through actual machining. From the result of cutting tests, it was verified that the proposed method can more accurately estimate the cutting force with a machine tool having multi-inertia dynamic system.

Tool life and heat-affected area in subsurface should be controlled with the cutting temperatures in terms of product quality. The cutting temperature depends on the tool material as well as the cutting conditions and the tool geometries. The thermal properties of the tool material, the major factor in control of the cutting temperatures, should be studied to employ the appropriate cutting tool. This paper discusses the effect of the thermal properties on the cutting temperatures in cutting simulations of drilling. In the cutting model, three-dimensional chip flow in drilling is interpreted as a piling up of the orthogonal cuttings in the planes containing the cutting velocities and the chip flow velocities, in which the chip flow direction is determined to minimize the cutting energy. Then, the cutting temperature distribution is simulated with the heat conductions, the thermal convections, and the heat generations in the discrete elements of the tool, the chip, and the workpiece. The heat generations on the shear plane and the rake face are estimated as the energy consumption associated with the stress distributions given by the cutting force simulation. To design tool materials, the cutting temperatures are analyzed for some cutting tools with virtual thermal properties, which are associated with the volume fraction of material components.

Combining topology optimization and advanced manufacturing techniques can generate and fabricate parts of complex geometry. In this paper, we present density gradient based formulations for hybrid casting and additive manufacturing in the density-based topology optimization. A location-based Heaviside function is introduced to represent subregions divided by the parting surface. The optimized part in the subregions can be fabricated with casting or additive manufacturing. Constraints related to these two manufacturing techniques are imposed through volume integrals of the location Heaviside function projected density-gradient. Due to the smoothed parameterization of the parting surface, the developed approach can optimize the part and the parting surface simultaneously. Numerical examples are presented to demonstrate the validity and efficiency of the proposed approach.

In order to improve the quality of parts machining without depending on skilled operators, the standardization of setting operation is desired. Thus, it is necessary to extract and generalize the skills related to the setting operation. In this study, eye tracking technology is adopted to monitor the movement of the operator's interest and to obtain the gaze data during the setting operation. An AR (Augmented Reality) system is developed to support the setting operation for an inexperienced operator by indicating recommended clamping way of workpiece in turning. The important points are estimated from the gaze data of skilled operators and are marked to pay attention to inherit the skills of the setting operation in the AR system.

To appropriately proceed with a machine design, it is important to define the part shape while confirming the thickness distribution throughout the part as needed. In this study, we developed a novel method to visualize the thickness distribution in a solid object using a translucent point cloud. In this method, numerous points are randomly placed inside a solid object, then the volumetric thickness value at each point is computed. Efficient computation is realized using hierarchical axis-aligned bounding boxes to manage the maximum inscribed spheres inside the solid and parallel processing technology of a graphics processing unit (GPU). To improve the performance, a cell structure based on an orthogonal grid is introduced to efficiently detect the axis-aligned bounding boxes (AABBs) commonly indicated by close points and store them in high-speed shared memory. By rendering the translucent points that are color-coded according to their thickness values, the thickness distribution inside a solid object can be represented in an easy-to-understand manner. Numerical experiments are performed to verify the effectiveness of the proposed algorithm.

In manufacturing processes, the time and cost required for setting up numerically controlled machine tools greatly affects production efficiency. In the present study, we developed an algorithm for analysis of the tool path during scanning processing, which is a method often used for curved surfaces such as dies and molds. The algorithm is implemented on a graphics processing unit to achieve high speed. The method is based on a standard triangular mesh, which allows highly versatile modeling in three-dimensional computer-aided design. Using this method, it was possible to determine the optimum tool diameter for different approach positions in a reasonable amount of time.

The construction of approximators for simulations such as the finite element method using machine learning has the problem of a conflict between the reduction of training data generation time and approximation accuracy. Hybrid neural networks have been proposed as a fast approximator for simulations to solve this problem. Training approximators construct it with simple perceptrons with linear activation functions created based on deductive knowledge that can be approximated with less data. However, in the simulation of complex structures, only a limited number of phenomena can be modeled with deductive knowledge. Therefore, there are errors in the predictions. Therefore, a correction approximator has been created by having a neural network learn the errors in the predictions. This approximator is predicted with high accuracy by combining the results of these two approximators. This paper proposes a neural network with a structure that integrates these approximators. Unlike HNN approximators, which prioritize linear approximators, INN optimizes the sharing ratio between linear and nonlinear approximators through learning. This method allows INN to improve the accuracy of approximators and reduce the conflict between the number of training data and accuracy.

The flexibility of transportation using autonomous mobile robots (AMRs) in manufacturing factories can be improved by introducing a manual operation mode, in which the operator interacts with the robot body. In the previous study, a manual operation mode was implemented with a force-sensorless power-assist function according to impedance control based on the robot body sway produced by the operator's touching force on a stationary AMR. In this paper, a mode-switching algorithm is proposed to switch from autonomous navigation to manual operation for an autonomous driving AMR. Control-mode switching is achieved by comparing the observed sway with the estimated behavior based on the mathematical model of the AMR to distinguish the sway due to the operator's touching force from those due to acceleration/deceleration generated during autonomous navigation. The experimental results demonstrate that the control mode can be successfully switched to the manual operation mode using the proposed algorithm without any additional sensors.

We focus on a method to generate the order of parts to assemble products considering the work difficulty in the parts assembly process. The quantitative evaluation value of work difficulty is created by using motion analysis method time measurement for the parts assembly process, and a mathematical model is constructed to generate the order of parts to assemble. In this study, we explain the characteristics of the quantitative evaluation value of the work difficulty and the mathematical model developed to generate the order that minimizes the total work difficulty. Numerical experiments are performed to evaluate the performance of the mathematical model, and the results obtained are compared with the resultant operation time obtained from actual parts assembly work.

Cold spray is a promising area of additive manufacturing concerning the repair of parts with complicated surface geometry in various application domains. However, as currently cold spray is mainly human-operated, it is limited in both precision and efficiency. Integration of this technique with a robot can unleash its potential with improvements to both limitations. A crucial component in a robotic system of this nature is part inspection to identify and quantify defect. We propose the use of a laser profiler incorporated into the robot arm to facilitate part surface examination throughout cold spray repair. of dents and wear spots. Focusing on a selectable area of a part surface, the laser profiler automatically captures high precision (1µm repeatability) overlapping point clouds of the surface and its neighborhood which are stitched together through ICP to return a single point cloud detailing the inspection area. This point cloud is subsequently fed through a suite of comparison algorithms along with the part’s CAD model to isolate and quantify areas of defect. An unworn area comparison to the corresponding CAD model is used to isolate the noise associated with the motion of the robot end-effector. This is achieved by adjusting the point cloud offset to the CAD model through a Möller-Trumbore ray-triangle intersection algorithm. A geometric dissimilarity approach is then applied to isolate points that correspond to surface defects. This approach inspects Euclidean distance and Gaussian curvature at each point in the captured scan with respect to the CAD model and automatically separates points with high disagreement. Implementation results on benchmark specimen demonstrate the high accuracy and robustness of the proposed system.

Directed energy deposition (DED) is a type of additive manufacturing (AM) that uses a metal powder as the material and a laser as the heat source. A laser beam is irradiated on a metal surface to create a melt pool, and metal powder is sprayed into the pool for modeling. For high-accuracy, high-efficiency, and high-quality molding in DED, the stand-off distance (SOD) between the head nozzle and melt pool must be maintained at an optimal value. However, owing to the many complex parameters of DED, the height of the modeled object is unstable, and the SOD cannot be maintained at an optimal level.

Three systems were developed in this study. The first system measures the SOD of the Nth layer in-process using a CMOS camera and calculates the build height of the N+1th layer. The second generates the modeling paths according to the calculated N+1th build height. In addition, it optimizes the modeling direction and pattern of path according to the shape. The third generates NC programs in real time based on the generated modeling paths, and models the object. The effectiveness of the overall system was confirmed through verification experiments.

Additive manufacturing (AM) allows objects to be designed and formed directly from three-dimensional data. Directed energy deposition (DED), an AM method, can produce near net shape parts, which can reduce lead time and material waste, resulting in lower costs. From the growing interest in vehicle electrification, usage of AM technology for functional materials such as soft magnetic materials has been attracting attention. Performance indicators of soft magnetic materials are high saturation polarization, high susceptibility, and low coercivity. These properties depend on microstructure and crystal texture. In this research, experiments were carried out to find ideal process parameters of DED using two soft magnetic materials, FeSi and FeCoV. Crystal orientation of both materials showed anisotropic properties with columnar grains in the build direction. Six process parameters, laser power, feed rate of nozzle head, dwell time, rotational speed of disc in powder pod, powder supply rate, and carrier gas speed were adjusted to perform a screening test. It was found that dwell time and powder supply rate have a large influence on the shape of deposit. Also, higher deposit density is achieved by higher laser power and lower feed rate. Consequently, optimal process parameters of both FeSi and FeCoV deposits were obtained.

Some additive manufacturing processes include melting and solidifying the material for forming the target shape by layer by layer. Rapid heating and cooling techniques have been discussed by many researchers until now; however, these processes are often dealt with as different challenges. For example, many researchers used high-power laser for rapid melting, and developed cooling device for rapid solidification to achieve high deposition efficiency, whereas heating and cooling are totally opposite problems leading much energy consumption when they are discussed separately. By discussing these contradicted challenges at the same time, it is possible to find a new challenge for avoiding unnecessary heat energy supply onto the deposition point to achieve high efficiency. This discussion is also important to use the materials with low melting temperature in additive manufacturing, which often requires a high-performance cooling system. This study proposes an additive manufacturing process applying supercooling phenomenon. Supercooling is a metastable state in which a material keeps liquid phase even under its melting temperature, and immediately solidifies by applying disturbance. This study evaluates the controllability and applicability of supercooling on the deposition process by dropping droplets of supercooled H₂O and gallium onto the baseplate.

In recent years, with the spread of electric vehicles, development of high-performance mechanical parts has been required. With respect to the manufacturing process, metal Additive Manufacturing (AM) such as Directed Energy Deposition (DED) is attracting attention in automobile industry. Since the DED process can deposit molten material onto the specific surface, where it solidified, it is regarded as a promising coating process to enhance the hardness and surface functionality of mechanical parts. From this perspective, a gear coating process using DED is proposed in this research to strengthen gear surface and its basic study is conducted. Concretely, the coated material quality and mechanical properties like hardness are evaluated.

Surface defect detection is a essential feedback for quality control in manufacturing processes. Existing methods used for defect detection are mainly based on image-based approaches that is subject to the disturbances in light exposure or surface reflectivity, especially for the textured surfaces. In this paper, a new defect detection method is presented to detect surface defects using laser scanning point cloud data, which contain the full profiles of the surface in 3D space. The developed method converted the 3D point cloud data to surface normal vectors to normalize the scale of the geometry, and then extracted defect-induced geometric features from the surface normal via Gabor filter. The feasibility of the method has been tested in case studies for detecting different types of defects on the textured surfaces through simulations. The results show the effectiveness and robustness of the proposed method, and demonstrated improved performance compared to the conventional method based on Region Growing Segmentation algorithm.

Equipment failures lead to unexpected downtimes that often result in a significant capital and productivity loss. A condition monitoring system (CMS) is critical in alleviating such equipment downtime and to ensure reliability by providing real-time analytics on condition monitoring (CM) data. However, many small and medium enterprises find CM data inaccessible due to the requirement of expensive hard-wired data acquisition system and the difficulty to install the additional hardware and sensors within the existing plant set-up. In this regard, Internet of Things (IoT) brings a new opportunity to CMS, increasing the system's ability to collect and analyze data with more flexibility. This paper presents a proof-of-concept prototype of an IoT-enabled CMS that utilizes edge-devices to remotely collect various measurements from industrial equipment and carry out real-time CM. Moreover, the system leverages a cloud, where history of machinery health state is stored. The cloud also enables real-time access of machinery health status either on demand or push notifications. This cloud service can be further expanded to facilitate the integration of advanced data analytic tools for predictive maintenance. The effectiveness of the implemented system is presented by demonstrating remote data acquisition and spectral analysis of CM signals from a rotating machinery test-rig.

The stereolithography process has been invented to cure liquid acrylic resin into optical lenses with the designated profiles that deliver desired optical performance. Camera sensors are used online to collect a series of high-dimensional images, from which the profile of the optical lens can be extracted and characterized. The state-of-the-art computer vision and image processing methods fall short in effectively extracting the profile from the noisy image background. This paper presents a generic methodology that decomposes an in-process image into three components, i.e., background, profile, and noise, according to their variant pixel intensities. Such intensity variance is the result of the difference between the refractive index of the cured optical lens and that of the surrounding curable resin. The physical properties of individual image components are explicitly formulated in a regularized tensor decomposition model and represented by the mathematical properties of the corresponding tensor components. An ADMM-based algorithm is developed to optimally estimate the model parameters and accurately extract the profile component. The advantages of the proposed method are demonstrated by a real-world case study.

Multivariate process monitoring using sensory data has become increasingly critical for quality control of industrial processes and equipment. The high sampling frequency of sensors generally results in strong autocorrelation within the data. However, extant research on multivariate process monitoring is typically based on the independence assumption. In this study, we present a framework for autocorrelated multivariate process monitoring and study various monitoring algorithms. Specifically, the conventional vector autoregressive (VAR) model, and emerging deep learning models such as the long short-term memory (LSTM) model and the LSTM-based autoencoder model are investigated. The monitoring performance of the different models are compared under different shift patterns based on the metrics of average run length and average computational time. The simulation results demonstrate that the residual-space monitoring scheme of the LSTM autoencoder model is effective in achieving the smallest average run length for mean shifts, while the VAR model and the LSTM model are better for monitoring correlation changes. The average computation time of the LSTM autoencoder-based model is relatively greater than other models, which needs to be further explored to accommodate real-time monitoring scenarios.

Five years ago the US Air Force Office of Scientific Research (AFOSR) issued a call for developing an autonomous material experimentation platform, where the system is able to execute, evaluate, and plan new experiments iteratively with minimal human intervention. Before and since then, there has been active research to accomplish this goal set forth in the AFOSR call. In this paper, we discuss the landscape of the existing autonomous experimentation systems with a focus on their intelligent control capability (i.e., the “planner”). A commonly used mechanism for planners is based on Bayesian optimization (BO) and, within the BO frame, a widely used acquisition function is the expected improvement criterion. To address autonomous systems' ability to handle surprising observations, we present our ongoing work in exploring a surprise-based BO method. We apply the method retrospectively to an autonomous experimental 3D printing system at the Air Force Research Laboratory and discuss some interesting observations.

An integral part of information-centric smart manufacturing is the adaptation of industrial robots to complement human workers in a collaborative manner. While advancement in sensing has enabled real-time monitoring of workspace, understanding the semantic information in the workspace, such as parts and tools, remains a challenge for seamless robot integration. The resulting lack of adaptivity to perform in a dynamic workspace have limited robots to tasks with pre-defined actions. In this paper, a machine learning-based robotic object detection and grasping method is developed to improve the adaptivity of robots. Specifically, object detection based on the concept of single-shot detection (SSD) and convolutional neural network (CNN) is investigated to recognize and localize objects in the workspace. Subsequently, the extracted information from object detection, such as the type, position, and orientation of the object, is fed into a multi-layer perceptron (MLP) to generate the desired joint angles of robotic arm for proper object grasping and handover to the human worker. Network training is guided by forward kinematics of the robotic arm in a self-supervised manner to mitigate issues such as singularity in computation. The effectiveness of the developed method is validated on an eDo robotic arm in a human-robot collaborative assembly case study.

Efficient and robust task planning for a human-robot collaboration (HRC) system remains challenging. The human-aware task planner needs to assign jobs to both robots and human workers so that they can work collaboratively to achieve better time efficiency. However, the complexity of the tasks and the stochastic nature of the human collaborators bring challenges to such task planning. To reduce the complexity of the planning problem, we utilize the hierarchical task model, which explicitly captures the sequential and parallel relationships of the task. We model human movements with the sigma-lognormal functions to account for human-induced uncertainties. A human action model adaptation scheme is applied during run-time, and it provides a measure for modeling the human-induced uncertainties. We propose a sampling-based method to estimate human job completion time uncertainties. Next, we propose a robust task planner, which formulates the planning problem as a robust optimization problem by considering the task structure and the uncertainties. We conduct simulations of a robot arm collaborating with a human worker in an electronics assembly setting. The results show that our proposed planner can reduce task completion time when human-induced uncertainties occur compared to the baseline planner.

The need to increase the flexibility of production lines is calling for robots to collaborate with human workers. However, existing interactive industrial robots only guarantee intrinsic safety (reduce collision impact), but not interactive safety (collision avoidance), which greatly limited their flexibility. The issue arises from two limitations in existing control software for industrial robots: 1) lack of support for real-time trajectory modification; 2) lack of intelligent safe control algorithms with guaranteed collision avoidance under robot dynamics constraints. To address the first issue, a jerk-bounded position controller (JPC) was developed previously. This paper addresses the second limitation, on top of the JPC. Specifically, we introduce a jerk-based safe set algorithm (JSSA) to ensure collision avoidance while considering the robot dynamics constraints. The JSSA greatly extends the scope of the original safe set algorithm, which has only been applied for second-order systems with unbounded accelerations. The JSSA is implemented on the FANUC LR Mate 200id/7L robot and validated with HRI tasks. Experiments show that the JSSA can consistently keep the robot at a safe distance from the human while executing the designated task.

In this paper, we propose a hybrid convolutional neural network architecture that can identify the melt pool depth (MPD), a melt pool parameter that is strenuous to extract with conventional sensing techniques, as a continuous value based on 2-dimension melt pool images in powder bed fusion (PBF) additive manufacturing. The melt pool images and their corresponding geometric parameters were collected from finite element methods. The average coefficient of correlation between the model prediction value and the true label value by five-fold cross validation is 0.980, demonstrating the effectiveness of the proposed architecture.

Roll-to-roll(R2R) peeling is an innovative method that transfers flexible electronics and 2D materials from the flexible substrate where they are grown to the end-use substrate. This process enables the full potential of R2R 2D material fabrication methods in a continuous, high-throughput, and environment-friendly manner. During the R2R peeling process, the device patterning causes periodic changes in the adhesion energy between the device and substrate. This periodic disturbance can degrade the quality of the final product if not properly controlled. Current control methods used for the R2R peeling process do not explicitly reject the periodic disturbance. It is therefore desirable to develop a controller that is capable of performing periodic disturbance rejection. This paper presents a model-based repetitive controller that integrates a frequency estimation of the disturbance into the R2R peeling control to maintain the optimal peeling process performance. A linear estimator using system identification techniques is employed. The simulation results show that the developed controller achieves better R2R process performance when compared to a conventional model-based controller.

A five-axis machining center is known for its synchronous control capability, allowing complicated three-dimensional surfaces, such as propellers and hypoid gears, to be quickly created. We aimed to maintain the feed speed vector at the end milling point by controlling two linear axes and a rotary axis with a five-axis machining center to improve the machined surface quality. In previous research, we suggested reducing the shape error of machined workpieces by applying the proposed method, which uses a parameter (referred to as precedent control coefficient in this research) to reduce the differences in the servo characteristics of the three axes in the machining method. Moreover, to maintain the feed speed vector at the end milling point when machining complex shapes, a rapid velocity change in each axis is required, causing inaccuracy owing to torque saturation. In this research, to reduce the shape error while avoiding torque saturation when movement has high angular velocity, we developed a theoretical method to obtain the most suitable precedent control coefficient of each axis by using a block diagram that considers torque saturation. Therefore, both shape error reduction and torque saturation avoidance can be realized by using the proposed method.

Dual servo systems are widely used as ultra-high-speed and accurate positioning mechanisms for large work-space precision positioning applications such as in ultra-high-speed laser machining (cutting). Dual servo systems consist of serial combination of coarse and fine servo drives yielding a redundant servo system. The coarse drive carries the fine drive and provides large-stroke positioning whereas fine drive undertakes ultra-high-speed small-stroke positioning of the laser system. This paper proposes a novel smooth trajectory generation algorithm for such redundant dual-drive servo systems. The proposed algorithm presents a novel interpolation approach, which continuously interpolates linear laser cutting paths while blending them smoothly to realize a non-stop continuous motion. The interpolation error is controlled within user-set limits to realize accurate laser positioning. Distributed synchronous motion commands for coarse and fine axes are generated to operate them strictly within their stroke limits. Furthermore, the proposed approach is based on convolution (filtering) which allows real-time implementation of the command generation approach. Simulation results show that the proposed algorithm can generate the synchronous motion trajectories for the dual-drive system with analytically control corner errors and within kinematic limits.

Worm-like robots are widely used in confined environments for special operations, such as pipeline inspection and survivor rescue. With the retrograde peristalsis wave as the fundamental locomotion mechanism, the worm-like robot may exhibit undesired backward slippage due to the failure of the anchoring segments, inducing the loss of locomotion efficiency. The backward-slippage phenomenon is closely related to robot dynamics and the working environment. To solve this problem, a prerequisite is used to identify the inclination angle of the pipeline. To this end, in this research, Inertial Measurement Units (IMU) are embedded into the worm-like robot for motion and environmental perception. Through attitude calculation and filtering algorithm, the real-time angle of the robot can be determined, and the changes in the inclination angles of the pipeline can be perceived. The effectiveness of the proposed solution is verified through experiments. This research greatly expands the perception capability of worm-like robots and provides a basis for feedback control to reduce slippage effects.

State perception of a worm-like robot in an unstructured environment is a prerequisite for precise locomotion control. One of the current mainstream methods is to use inertial measurement units (IMU). However, due to sensor drift and the existence of uncertainty in the environment, the state estimator errors will accumulate. To solve this problem and to make it applicable to worm-like robots, the IMU data of the robot segments under different locomotion gaits are experimentally acquired and examined. Based on the variance threshold, the acceleration, and the angular velocity of the robot segment, a sliding window multi-threshold zero-velocity update (ZUPT) method is proposed. This method can solve the problem of pseudo-zero velocity interval and pseudo-motion interval induced by gait switching and sensor cumulative error under a single judgment condition, thereby improving the accuracy of the IMU in the perception of the state worm-like robots.

Many underground shopping streets are located around terminal stations and the majority of passersby use them as a route for transferring between transportation systems, etc. Therefore, placing stores that attract the interest of passersby along their route will lead to an increase in the number of customers. However, it is difficult to determine the layout of stores in an underground shopping streets because people who have various purposes pass through and there are many stores. In this study, simulation and optimization methods are combined to determine the store layout. An agent simulation is constructed using the measured human flow data and used to evaluate the fitness of the optimization method. Computer experiments will be conducted to verify the effectiveness of the proposed method.

Articulated Arm Coordinate Measuring Machines (AACMM) have been increasingly used in the manufacturing industry in recent years because of its portability and flexibility in measurement. However, the measurement accuracy of AACMMs is often significantly worse than conventional CMMs. To improve the measurement accuracy of AACMM, the model-based compensation has been studied. The Denavit-Hartenberg (D-H) model, which is widely used in robotics field, only contains position and orientation errors of the rotary axis average lines. In this study, we propose a new kinematic model of AACMM with the angular positioning deviation “error map” of rotary axes. The angular positioning deviation is modelled as a function of the rotary axis angular position. A method to identify the angular positioning deviation of rotary axes is also proposed by constraining the stylus position at various locations over the workspace. A machine tool, with a magnetic sphere socket installed on its spindle, is used to position the stylus sphere at the prescribed locations.

For automated control of the bucket position and orientation of excavator, it is necessary to measure the 3D position of the bucket tip in real time. For its estimation by measuring the link orientations, it is necessary to accurately identify in advance the kinematic parameters of the excavator, such as the length of the boom, arm and bucket. In this study, we propose an identification method of excavator kinematic parameters by measuring the arm top position using a laser tracker. The estimation performance of the arm top position, based on the identified kinematic parameters and the link orientations measured by inertial measurement units (IMUs), is experimentally investigated.

To improve the absolute positioning accuracy of a six-axis industrial robot, our group has proposed a novel kinematic model containing not only Denavit-Hartenberg (D-H) parameters but also the bidirectional angular positioning deviations of each rotary axis. This research presents a new scheme to identify this model by using a sweeping laser measurement system. As the first step, this paper only presents the identification of the D-H parameters. A sweeping laser measurement system measures the displacement of the robot end effector in the direction normal to the sweeping laser plane at various positions on it. The main advantage of this measurement method is significantly lower instrumental cost compared to other devices like a laser tracker.

This paper introduced n-step hybrid flow-shop scheduling (nHFS) of product-mix Make-to-Order production with individual job's loading condition and delivery dates. The ability sometimes depends on the loading rate of a batch process and usable resource constraints due to a job specification in an actual manufacturing process. Based on the nHFS solution proposed by J.N.D Gupta, et.al. (2002) (hereby, n-Gupta), three extensions are presented. First is the optimization of weighting parameters of multi-objective functions without domain knowledge implementation. Particularly, higher performance ensemble methods of Bayesian optimization based on Gaussian Process regression are applied. Second is the appropriate product-mix batch loading considering restricted usable resources. The third is the extension of the multi-objective function to consider the flow balance and capacity distribution between the flows. Numerical evaluation results in a better performance than job-shop scheduling and actual companies' plan, in almost all of the delivery satisfactions, makespan, batch loading rate, and so on.

Grey wolf optimization (GWO) algorithm is a swarm intelligence optimization technique that is developed by Mirjalili [1] to mimic the hunting behavior and leadership hierarchy of grey wolves in nature. It has been successfully applied to many real site applications. However, there is still an insufficiency in the GWO algorithm regarding its position-updated equations and fewer operators, which is good at exploitation but poor at exploration and easy to trap in the local optima.

Oppositional-based learning (OBL) is a new concept, which has attracted many research efforts in the last decade, some optimization methods have already used the concept of OBL to improve their performance [2].

This paper presents an efficient algorithm, namely, Enhanced Mutation Oppositional-based learning GWO (EMOGWO) based on Quasi Oppositional-based learning (QOBL) and Topological Oppositional-based learning (TOBL) with some parameter adjustment to balance between exploration and exploitation.

The experiments were executed 28 widely used benchmark test functions with various features. The results reveal that the proposed algorithm shows better or at least competitive performance against other compared algorithms in terms of exploration and exploitation.

The volumetric accuracy of a machine tool generally changes with time. Its periodic check, performed at a user's site in a semi-automated manner, can be a key to ensure sufficient volumetric accuracy in a long term. The multilateration measurement, performed by using a tracking interferometer (laser tracker), can measure all the error motions of three linear axes. However, its higher instrumental cost prevents its wider implementation. This paper presents a prototype design of a tracking interferometer with a single rotary axis only. Since higher assembly accuracy is not required, it can be constructed with much lower cost. By using this instrument, a single tracking test is performed with two linear axes, in addition to direct measurement of their linear positioning deviations. An algorithm to identify 2D error motions of the two linear axes is presented. Experimental demonstration is presented to investigate its measurement accuracy.

A large magnetic field must be generated by a recording medium and applied to a magnetic sensor to improve the signal-to-noise ratio of a magnetic encoder. A method for calculating the magnetic field generated by various recording media is required to efficiently design a magnetic encoder with a large applied magnetic field. Therefore, the purpose of this study is to develop a simulation method to calculate the magnetic field applied to a magnetic sensor in magnetic encoders using various recording media. Two types of samples are prepared. One is a coating-type magnet recording medium, which mainly contains Fe-Co alloy particles. The other is a Fe-Ni alloy sheet medium. The finite element method and medium magnetization model are used to consider recording conditions and the influence of the demagnetizing field of the recording medium. The actual recording medium is magnetized under the same recording conditions as those used in the simulation. The amplitude of the detection signal due to the applied magnetic field is measured. As the magnitude of the recording current is expected to strongly affect the amplitude of the applied magnetic field, the simulation and experiment are performed with several different recording currents. The errors between the simulated and measured amplitudes for the coating-type and alloy sheet recording media are 3 % and 7 %, respectively. This indicates that the proposed simulation method accurately calculates the amplitude of the applied magnetic field. In addition, the simulation and measurement results show similar effects of the recording current on the amplitude. This shows that the proposed method can predict the optimum value of the recording current. Using the simulation method, an optimum combination of material and recording current can be predicted more accurately to obtain the larger amplitude. The simulation method can also be used for analysis of self-demagnetization due to shape. These contribute to the efficient design of magnetic encoders.

We have been developing a surface topography measurement system using laser inverse scattering with an optical frequency comb as a light source to measure microtopography of several nanometers to several micrometers in size with high accuracy and over a large area. The laser inverse scattering measures the surface topography by retrieving the phase distribution of the scattering light when a plane wave is incident on the sample surface. However, the dynamic range of this measurement is limited to sub-wavelength order. We propose a method to extend the dynamic range of the laser inverse scattering by using the retrieved phase at several different wavelengths. The method estimates the surface topography from the inversely proportional relationship between the phase and wavelength of light propagating in a fixed optical path. The phase distribution at several different wavelengths can be obtained by using an optical frequency comb as the light source of the laser inverse scattering.

We analyzed the scattering electric field at a sine wave shape surface with a height of 100 to 1000 nm, and simulated the surface topography measurement using the retrieved phase distribution at each wavelength. The simulation results show good agreement with the surface topography of the measurement target, confirming the accurate surface topography measurement and dynamic range extension with the proposed algorithm.

Released during 3-dimensional (3D) printing polymeric parts and many other industrial processes, the volatile organic compounds (VOCs) are mixtures that can pollute air quality and harm human health. Particularly, VOCs dominate in causing sick-building syndrome symptoms. Selective laser sintering (SLS) is a popular 3D technique, where the elevated temperature and high-energy laser beam tend to cause more detrimental VOC emission than other 3D printing processes. Though previous work reported to model emission in some 3D printing processes, the emission modelling and prediction in SLS has remained largely unexploited. This work proposes a mass-transfer and experiment-data based approach to model and predict VOC emission in SLS. We monitored the VOC emissions from polyamide 12 in 6 cases with varying printing parameters. With emission curve segmentation based on the physical printing process, we applied the mass-transfer single- and multiple-layer model to simulate the VOC emission. Model matching subsequently yields the suitable model type and node number in each printing stage. Results showed the proposed method has an average accuracy of 85.32% matching to the experiment results.

Resistance Spot Welding (RSW) has been an important welding process in many industries, including automotive and aerospace, given its great automation when combined with robots and electrical control systems. But real-time RSW quality inspection and assurance have been a challenge on shop floors. This paper presents an interpretable data-driven modeling method, upon neural networks and game theory-based Shapley values, for efficient and also interpretable prediction of weld quality metrics (e.g., nugget diameter, thickness, HAZ diameter) in RSW. The interpretation of the model predictions can contribute to the physical understanding of: 1) how would the weld quality vary under process uncertainties, e.g., varying sheet fit-up conditions; 2) how would in-situ sensing (e.g., resistance, displacement) benefit characterization of process uncertainties; 3) what would be the contributions of individual process parameters (e.g., welding current, time) and different sensing measurements to the prediction of weld quality. Through analyzing experimental data, these questions can be quantitively answered by the interpretation, which hence leads to a better understanding of RSW process dynamics and defect occurrences towards improved process control and quality assurance.