This paper develops a unified framework for training and deploying deep neural networks on the edge computing framework for image defect detection and classification. In the proposed framework, we combine the transfer learning and data augmentation with the improved accuracy given the small sample size. We further implement the edge computing framework to satisfy the real-time computational requirement. After the implement of the proposed model into a rolling manufacturing system, we conclude that deep learning approaches can perform around 30–40% better than some traditional machine learning algorithms such as random forest, decision tree, and SVM in terms of prediction accuracy. Furthermore, by deploying the CNNs in the edge computing framework, we can significantly reduce the computational time and satisfy the real-time computational requirement in the high-speed rolling and inspection system. Finally, the saliency map and embedding layer visualization techniques are used for a better understanding of proposed deep learning models.

In-situ inspection has drawn many attentions in manufacturing due to the importance of quality assurance. With the rapid growth of additive manufacturing technology, the importance of in-line/in-situ inspections has been raised to a higher level due to many uncertainties that could occur during an additive printing process. Given this, having accurate and robust in-situ monitoring can assist corrective actions for a closed-loop control of a manufacturing process. Contact 3D profilometers such as stylus profilometers or coordinate measuring machines can achieve very high accuracies. However, due to the requirement for physical contact, such methods have limited measurement speeds and may cause damage to the tested surface. Thus, contact methods are not quite suitable for real-time in-situ metrology. Non-contact methods include both passive and active methods. Passive methods (e.g., focus variation or stereo vision) hinges on image-based depth analysis, yet the accuracies of passive methods may be impacted by light conditions of the environment and the texture quality of the surface. Active 3D scanning methods such as laser scanning or structured light are suitable for instant quality inspection due to their ability to conduct a quick non-contact 3D scan of the entire surface of a workpiece. Specifically, the fringe projection technique, as a variation of the structured light technique, has demonstrated significant potential for real-time in-situ monitoring and inspection given its merits of conducting simultaneous high-speed (from 30 Hz real-time to kilohertz high speeds) and high accuracy (tens of μm) measurements. However, high-speed 3D scanning methods like fringe projection technique are typically based on triangulation principle, meaning that the depth information is retrieved by analyzing the triangulation relationship between the light emitter (i.e., projector), the image receiver (i.e., camera) and the tested sample surface. Such measurement scheme cannot reconstruct 3D surfaces where large geometrical variations are present, such as a deep-hole or a stair geometry. This is because large geometrical variations will block the auxiliary light used in the triangulation based methods, which will resultantly cause a shadowed area to occur. In this paper, we propose a uniaxial fringe projection technique to address such limitation. We measured a stair model using both conventional triangulation based fringe projection technique and the proposed method for comparison. Our experiment demonstrates that the proposed uniaxial fringe projection technique can perform high-speed 3D scanning without shadows appearing in the scene. Quantitative testing shows that an accuracy of 35 μm can be obtained by measuring a step-height object using the proposed uniaxial fringe projection system.

In-situ layer-by-layer inspection is essential to achieving the full capability and advantages of additive manufacturing in producing complex geometries. The shape of each inspected layer can be described by a 2D point cloud obtained by slicing a thin layer of 3D point cloud acquired from 3D scanning. In practice, a scanned shape must be aligned with the corresponding base-truth CAD model before evaluating its geometric accuracy. Indeed, the observed geometric error is attributed to systematic, random, and alignment errors, where the systematic error is the one that triggers an alarm of system anomalies. In this work, a quickest change detection (QCD) algorithm is applied under a multi-resolution alignment and inspection framework 1) to differentiate errors from different error sources, and 2) to identify the layer where the earliest systematic deviation distribution changes during the printing process. Numerical experiments and a case study on a human heart are conducted to illustrate the performance of the proposed method in detecting layer-wise geometric error.

Recent measurement system developments have brought new opportunities to enhance the performance of quality control (QC) systems in manufacturing. Digital cameras and 3D optical scanners are among the advanced measurement systems that can represent an entire product’s surface. These data-rich environments have been widely used in automated surface defect detection. However, despite the fact that datasets from both digital cameras and 3D optical scanners can both represent a surface, the datasets are fundamental different and contain different information regarding a part’s surface. Extensive research efforts have been conducted on developing QC tools for each of these datasets individually, very few researches were focused on approaches to take advantage of both 3D point clouds (obtained from technologies such as 3D optical scanners) and digital images (obtained from digital cameras) at the same time. This paper proposes a hybrid approach for surface inspection of additive manufacturing parts through Multilinear Principal Component Analysis (MPCA). This proposed approach is demonstrated with additively manufactured parts, which shows the advantage of combining this information for surface classification.

With the latest advancements in three-dimensional (3D) measurement technologies, obtaining 3D point cloud data for inspection purposes in manufacturing is becoming more common. While 3D point cloud data allows for better inspection capabilities, their analysis is typically challenging. Especially with unstructured 3D point cloud data, containing coordinates at random locations, the challenges increase with higher levels of noise and larger volumes of data. Hence, the objective of this paper is to extend the previously developed Adaptive Generalized Likelihood Ratio (AGLR) approach to handle unstructured 3D point cloud data used for automated surface defect inspection in manufacturing. More specifically, the AGLR approach was implemented in a practical case study to inspect twenty-seven samples, each with a unique fault. These faults were designed to cover an array of possible faults having three different sizes, three different magnitudes, and located in three different locations. The results show that the AGLR approach can indeed differentiate between non-faulty and a varying range of faulty surfaces while being able to pinpoint the fault location. This work also serves as a validation for the previously developed AGLR approach in a practical scenario.

Basing on the laser-triangulation principle, a deep hole inner surface inspection system is developed in this research. Compared with existing inner surface measurement systems, this research takes assembly errors and refraction distortion into consideration and proposes a flexible laser plane calibration technique based on binocular vision that is easy to operate. The measurement system calibrated by proposed method performs well in the field test in which the maximum absolute error of the measured ring gauge diameter is 3μm. Experiment indicates that this system is accurate and suitable for inner surface measurement. Point cloud registration may be applied in the future to extend the system’s applicability.

In condition-based maintenance, preventive replacement threshold and inspection scheme play important roles in maintenance performance. Major research considers cost as the main objective for measuring maintenance performance; here the average cost per unit time is used as the only objective in a single-unit system. The intention of this study was to investigate that how the average cost per unit time varies through changing replacement threshold and inspection scheme, the aim is to simultaneously determine an optimal replacement threshold and inspection scheme. The heterogeneity of hazard rate over stages of the equipment entails less frequent inspections when the equipment is at a healthy condition and more frequent inspections when the equipment is at a progressively deteriorated condition. Therefore the proposed condition-based inspection in the present work is non-periodic. A method based on dynamic programming is developed in order to implement a condition-based inspection scheme; furthermore the threshold for preventive replacement and the inspection scheme are simultaneously determined and then the optimal average maintenance cost is obtained.

Non-Destructive Evaluation (NDE) of welded structures is essential in industry and manufacturing sectors. However, NDE techniques are limited when applied at high temperatures, which prevents usefulness for on-line real time inspection of welded joints. In this work, a high temperature (HT) inspection system was created utilizing Phased Array Ultrasonic Testing (PAUT), and tested on Friction Stir (FS) welded aluminum alloy joints. The system created in this work proves HT-PAUT is capable of determining defects during the welding process. Supplementing this work, a custom defect detection software was created to analyze S-Scan data to interpret when and where a defect occurs to provide defect indicator signals. These defect signals can be utilized for controlling the FSW process parameters to automatically correct if a defect is observed. The technology developed can be utilized as a platform for future automated welding processes and control for creating the next generation weld-NDE systems.

In most trendsetting industries like the aerospace, automotive and medical industry functionally critical parts are of highest importance. Due to strict legal requirements regarding the securing of the functionality of high-risk parts, both production costs and quality costs contribute significantly to the manufacturing costs. Thus, both types of costs have to be taken into consideration during the stage of technology planning. Due to the high variety of potential interactions between individual component properties as well as between component properties and manufacturing processes, the analysis of the influence of the manufacturing history on an efficient design of inspection processes and inspection strategies is extremely complex. Furthermore, the effects of inspection strategies and quality costs on the planning of manufacturing process sequences cannot be modeled to date. As a consequence, manufacturing and inspection processes are designed separately and thus a high cost reduction potential remains untapped. In this paper a new approach for an integrative technology and inspection planning is presented and applied to a case study in medical industry. At first, existing approaches with regard to technology and inspection planning are reviewed. After a definition of relevant terms the case study is introduced. Following, an approach for an integrative technology and inspection planning is presented and applied to the case study. In the presented approach the complex causalities between technology planning, manufacturing history and inspection planning are considered to enable a cost-effective production process and inspection sequence design.

In order to inspect the condition of micro milling cutter automatically and accurately in the on-line process, a dedicated micro milling cutter condition inspection system was established in this paper, which includes an on-machine inspection device and a controller. The key methods of the automatic dimension measurement and the cutting edge condition inspection for micro milling cutters are studied. The proposed methods can measure cutter diameter and clamping height, classify tool type as well as inspect cutting edge condition from both radial and axial direction. The experiments verify that the proposed methods and the developed inspection system can fulfill the needs of industrial applications.

Laser powder-bed fusion (L-PBF) is an additive manufacturing (AM) process that enables fabrication of functional metal parts with near-net-shape geometries. The drawback to L-PBF is its lack of dimensional precision and accuracy. The efficiency of powder fusion process in powder-bed AM processes is highly affected by process errors, powder irregularities as well as geometric factors. Formation of defects such as lack of fusion and over-fusion due to the aforementioned factors causes dimensional errors that significantly damage the precision. This paper addresses the development of an automated in-situ inspection system for powder-bed additive manufacturing processes based on machine vision. The results of the in-situ automated inspection of dimensional accuracy allows for early identification of faulty parts or alternatively in-situ correction of geometric errors by taking appropriate corrective actions. In this inspection system, 2D optical images captured from each layer of the AM part during the build are analyzed and the geometric errors and defects impairing the dimensional accuracy are detected in each layer. To successfully detect geometric errors, fused geometric objects must be detected in the powder layer. Image processing algorithms are effectively designed to detect the geometric objects from images of low contrast captured during the build inside the chamber. The developed algorithms are implemented to a large number of test images and their performance and precision are evaluated quantitatively. The failure probabilities for the algorithms are also determined statistically.

One of the major challenges in nanoscale manufacturing is defect control because it is difficult to measure nanoscale features in-line with the manufacturing process. Optical inspection typically is not an option at the nanoscale level due to the diffraction limit of light, and without inspection high scrap rates can occur. Therefore, this paper presents an atomic force microscopy (AFM)-based inspection system that can be rapidly implemented in-line with other nanomanufacturing processes. Atomic force microscopy is capable of producing very high resolution (sub-nm-scale) surface topology measurements and is widely utilized in scientific and industrial applications, but has not been implemented in-line with manufacturing systems, primarily because of the large setup time typically required to take an AFM measurement. In order to overcome this limitation, we have developed a single-chip-AFM-based inspection system where a wafer can be precisely and repeatably loaded into the setup and measurements can be taken in under 60 seconds. This inspection system consists of several single-chip AFMs integrated into a positioning stage to make measurements at multiple spots on a wafer at the same time. Each single-chip AFM is a MEMS device that is approximately 2 mm wide by 1 mm tall and is capable of scanning a 10 micron by 10 micron area. Thermal actuators in the MEMS device are used to do the scanning in both the x and y directions as well as to excite the z axis of the AFM so that it can be run in taping mode. Each AFM is attached to a flexure stage in the top plate of the inspection system so that the AFM can be precisely moved to the desired inspection location on the wafer. The flexure plate is coupled to the inspection plate using a kinematic coupling so that the flexure plate can be precisely located with respect to the inspection plate after each loading operation. In order to take a measurement, the flexure plate is removed from the inspection plate, a wafer is loaded into the inspection plate using an exactly constrained, passive alignment system, and the flexure plate is then placed back onto the inspection plate. This brings the AFMs back into contact with the surface that is to be measured and the AFMs can then start taking measurements without any additional alignment operations. The overall measurement procedure takes less than one minute, which is faster than most nanomanufacturing processes. This guarantees that the inspection step will not be the bottleneck in the manufacturing process.

The position and orientation deviation of workpiece caused by the fixing errors of the fixture influences the machining accuracy of workpiece. This paper presents a method of predicting the fixture induced error and its influence on the machining accuracy of features pattern. Firstly, the mathematical model for predicting fixing error caused by locators and locating surface geometrical errors and caused by clamping force and machining force is formulated. Then, based on the 3-2-1 positioning fixture, the influence of fixing errors on the machining accuracy of features pattern is analyzed after introducing the definition, inspection and evaluation method for the composite position error of features pattern. Finally, taking 4-holes pattern in rectangular distribution as an example, the process of predicting the composite positional error of features pattern is detailed and a simulation experiment is conducted. The results show the rightness and validity of this method.

It is widely observed that today’s engineering products demand increasingly strict tolerances. The shape of a machined surface plays a critical role to the desired functionality of a product. Even a small error can be the difference between a successful product launch and a major delay. Thus, it is important to develop measurement tools to ensure the quality and accuracy of products’ machined surfaces. The key to assessing the quality is robust measurement and inspection techniques combined with advanced analysis. However, conventional Geometrical Dimensioning and Tolerancing (GD&T) such as flatness falls short of characterizing the surface shape. With the advancements in metrology methodology utilizing digital holographic interferometry, large amount of surface data can be captured at high resolution and accuracy without changing platform or technique. This captured High Definition Data (HDD) enables the mining of more valuable information from machined surfaces that most current industry practice cannot achieve in a timely manner. Such new metrology system opens the torrent of observable events at plant floor and increases the transparency of machining processes. This presents great opportunities to characterize machined surface into a new level of details, which can be applied in production quality evaluation and process condition monitoring and control. This research work proposes a framework of a multi-scale surface characterization for surface quality evaluation and process monitoring. Case studies are presented to show how proposed metrics could be applied in surface quality evaluation and process monitoring.

Development of industry demands better performance of equipments and devices than ever. The property of material used to produce the equipments is the precondition to ensure the quality. As equipments are usually required to be integral during the quality inspection, non-destructive testing (NDT) plays an increasingly import role in modern industry quality control. Different NDT methods are introduced and analyzed. The mechanism of ultrasonic exciting is described. After the ultrasonic is excited, the vibration signal can be detected by transducer or optical method which are then illustrated. In the section of development of laser ultrasonic technology, contributions made by various researchers in theoretic development, experiment, simulation and application are introduced and the corresponding content of the researches as well. The conclusion and the outlook of laser ultrasonic technique is made at the last.

Research regarding tool wear in the machining of difficult materials is important because it is a significant indicator of process failure in terms of degradation of part quality, and the resulting high cost and increased process time. Prior researchers have investigated the effects of cutting parameters on tool wear and as a result, tool life has seen significant improvement. However, these studies are not concerned with tool flank wear during machining; they instead focus on tool flank wear after a certain amount of cutting distance. This study proposes a new method of predicting tool flank wear during machining that has the capability of suggesting tool failure without directly measuring the tool. For this purpose, a detailed set of experiments on end milling of titanium alloy Ti-6Al-4V was conducted and analyzed. Then, the resultant force output, which can be monitored during machining, was used to establish a predictive algorithm for tool flank wear. Using the increase in the resultant force as well as the total energy spent on the workpiece, it was shown that tool flank wear can be effectively predicted during machining and this can decrease the time spent on tool failure inspection and early tool change, increasing the throughput of the process.

Ever advancing sensor and measurement technologies continually provide new opportunities for knowledge discovery and quality control (QC) strategies for complex manufacturing systems. One such state-of-the-art measurement technology currently being implemented in industry is the 3D laser scanner, which can rapidly provide millions of data points to represent an entire manufactured part’s surface. This gives 3D laser scanners a significant advantage over competing technologies that typically provide tens or hundreds of data points. Consequently, data collected from 3D laser scanners have a great potential to be used for inspecting parts for surface and feature abnormalities. The current use of 3D point clouds for part inspection falls into two main categories; 1) Extracting feature parameters, which does not complement the nature of 3D point clouds as it wastes valuable data and 2) An ad-hoc manual process where a visual representation of a point cloud (usually as deviations from nominal) is analyzed, which tends to suffer from slow, inefficient, and inconsistent inspection results. Therefore our paper proposes an approach to automate the latter approach to 3D point cloud inspection. The proposed approach uses a newly developed adaptive generalized likelihood ratio (AGLR) technique to identify the most likely size, shape, and magnitude of a potential fault within the point cloud, which transforms the ad-hoc visual inspection approach to a statistically viable automated inspection solution. In order to aid practitioners in designing and implementing an AGLR-based inspection process, our paper also reports the performance of the AGLR with respect to the probability of detecting specific size and magnitude faults in addition to the probability of a false alarms.

The ASME Y14.5M standard has defined different types of tolerances that can be applied to a feature to achieve the required functionality. Each tolerance defines a zone within which the feature under inspection must lie. The conformance of the parts to these tolerances is checked by manual measurements or a CMM. But it has been observed that the measurements between different CMMs do not match. There are two generally accepted reasons for this discrepancy. The first one is the measurement uncertainty in CMM software. This problem was addressed by NIST by developing reference softwares for feature fitting algorithms. And the second one is the distinct choice of algorithms for fitting substitute feature to the data points measured from CMM. Feature fitting algorithms used in CMMs are often based on their mathematical convenience rather than the interpretation of definitions in the GD&T standard. Our research is focused on identifying that normative algorithm that is best to be used for each type of tolerance. Each normative algorithm is identified as the one to best represent the interpretation of geometric control as defined by the Standard and on the manual methods used for the measurement of a specific tolerance type.

A 38 MMSCF/D Cooper Bessemer Model LM-9 reciprocating compressor in hydrogen service at the Praxair Westlake LA facility has experienced notable particulate contamination within the feed gas. The particulates were believed to be caused by upstream piping corrosion; however, to definitely state the cause, the properties of the fluid existing in the five-stage compressor needed to be more fully understood. An OLI electrochemical simulation software was used for dew point prediction, determination of the condensed phase ionic equilibria, and corrosion rate prediction. These tasks were beyond capabilities of the site-licensed UniSim software, as presently configured. Specifically, the model was used to identify the dew point conditions (temperature, pressure) and properties of the condensed water (pH, corrosivity, dissolved O 2 , and chlorine speciation). Model results were compared with site inspection findings. Subsequently, recommended limits for chlorine and oxygen in the feed gas were established to improve long term compressor reliability.

The choice of fitting algorithm in CMM metrology has often been based on mathematical convenience rather than the fundamental GD&T principles dictated by the ASME Y14.5 standard. Algorithms based on the least squares technique are mostly used for GD&T inspection and this wrong choice of fitting algorithm results in errors that are often overlooked and leads to deficiency in the inspection process. The efforts by organizations such as NIST and NPL and many other researchers to evaluate commercial CMM software were concerned with the mathematical correctness of the algorithms and developing efficient and intelligent methods to overcome the inherent difficulties associated with the mathematics of these algorithms. None of these works evaluate the ramifications of the choice of a particular fitting algorithm for a particular tolerance type. To illustrate the errors that can arise out of a wrong choice of fitting algorithm, a case study was done on a simple prismatic part with intentional variations and the algorithms that were employed in the software were reverse engineered. Based on the results of the experiments, a standardization of fitting algorithms is proposed in light of the definition provided in the standard and an interpretation of manual inspection methods. The standardized fitting algorithms developed for substitute feature fitting are then used to develop Inspection maps (i-Maps) for size, orientation and form tolerances that apply to planar feature types. A methodology for Statistical Process Control (SPC) using these i-Maps is developed by fitting the i-Maps for a batch of parts into the parent Tolerance Maps (T-Maps). Different methods of computing the i-Maps for a batch are explored such as the mean, standard deviations, computing the convex hull and doing a principal component analysis of the distribution of the individual parts. The control limits for the process and the SPC and process capability metrics are computed from inspection samples and the resulting i-Maps. Thus, a framework for statistical control of the manufacturing process is developed.