Mechanical and Mechatronics Engineering
Permanent URI for this collectionhttps://uwspace.uwaterloo.ca/handle/10012/9912
This is the collection for the University of Waterloo's Department of Mechanical and Mechatronics Engineering.
Research outputs are organized by type (eg. Master Thesis, Article, Conference Paper).
Waterloo faculty, students, and staff can contact us or visit the UWSpace guide to learn more about depositing their research.
Browse
Browsing Mechanical and Mechatronics Engineering by Author "Bedi, Sanjeev"
Now showing 1 - 6 of 6
- Results Per Page
- Sort Options
Item Anomaly Detection in CNC Milling Machines using Transfer and Incremental Ensemble Learning of LSTM Autoencoder Networks(University of Waterloo, 2024-07-08) Li, Eugene; Bedi, Sanjeev; Melek, WilliamSince the industrial revolution, there has been a steady and continued effort to bring more automation and efficiency to the manufacturing process. Milling machines have long been a valuable tool to create precise parts quickly and effectively. CNC Milling machines have been the next evolution of automation in manufacturing and have allowed for complex parts to be produced quickly and with high precision. Although CNC milling machines are able to semi-autonomously produce parts effectively, they still require significant human interaction to operate. This human interaction is especially true since many tasks are completed in open loop configuration with little to no feedback. To try to address this issue there has been significant effort in the literature to develop systems to provide feedback to the machine controller. This work is often focused on detecting anomalies such as chatter, broken tools and other conditions that will impact the surface finish or machine health. A limitation of much of the current work is that it tends to be machine or material specific. These approaches developed in the lab often do not scale well to production as they require custom setups or complex machine dynamics to be studied. To overcome this problem, this thesis proposes a machine learning based solution that leverages deep learning to create a solution that can potentially be quickly and easily transferred to machines in production. In this thesis we demonstrate that by using simple accelerometers mounted on the spindle of a CNC milling machine, we can create an LSTM-Autoencoder to detect anomalies such as chatter. This feat is accomplished by creating an artificial neural network that is trained on sensor data from stable cutting conditions. This network aims to reproduce the original signal with as little error as possible. If the network is provided data from a stable condition it will reconstruct the signal with little error, but if it is presented data from an anomaly condition it will reconstruct with significant error, which indicates that an anomaly is present. In Chapter 4 we show that this approach can also be achieved by implementing what is known as transfer learning. In transfer learning we begin with a network that is trained on one source data set, and then transfer the knowledge to another target data set. We investigate under what conditions this is most feasible and demonstrate that we can train a network from a robust data set on one three-axis CNC machine and then transfer it to another three-axis CNC machine. We also demonstrate that this method works for both chatter detection and broken tool detection. In Chapter 5, we introduce an incremental learning method based on ensemble learning. This approach takes the LSTM-Autoencoder trained previously as a strong learner and has weak learners continually learn as new data is made available. This approach is shown to have comparable results to a large network trained from scratch and improves the performance of a system trained with transfer learning. Taking these transfer learning and incremental learning algorithms, we extend the approach to anomaly detection for five axis CNC milling machines in Chapter 6. This is accomplished by introducing a stacked ensemble learning approach by transferring the encoder from the three axis CNC anomaly detection algorithm and then combining it with an encoder and decoder that is trained from the target data. Incremental learning is then integrated by adding weak learners to this strong learner. These weak learners allow the network the ability to improve the performance of the system to be comparable to a network trained from scratch with a fraction of the data. Lastly, we demonstrate how these approaches can be used for multi-class prediction in Chapter 6. In this chapter, we use the LSTM-Autoencoder to perform dimensionality reduction. We then use this dimensionality reduced output and apply a one-versus-all SVM classifier and Platt scaling to obtain a probabilistic prediction of the classes of interest. This approach allows us the ability to both detect and differentiate cutting with broken tools and chatter conditions. The approaches presented in this thesis demonstrate that this proposed approach is capable of not only detecting chatter in a specific lab setting, but can potentially be used to detect multiple anomalies across a variety of machines and materials. This allows users to potentially scale these approaches to many machines quickly with minimal setup and minimal configuration.Item Application of Stereo-Depth Camera and Alignment Algorithms for Part Monitoring During Machining(University of Waterloo, 2023-11-20) Delattre, Samuel; Bedi, Sanjeev; Mann, StephenA stereo-depth camera is proposed to be used in conjunction with fiducial markers on a calibration plate and a fine-tuning alignment algorithm for part monitoring in a CNC machine. Together, a selected pyramid-shaped part within the machine could be monitored. The position, orientation, geometry and surfaces of the pyramid part are measured and compared with the pyramid’s desired model. This system can monitor the position and geometry within 1mm of accuracy, orientation within 1 degree of accuracy, and surface fitting within 2mm of accuracy, which closely aligns with the advertised accuracy of the stereo depth camera. While the accuracy is not enough to be confident if the part was machined to an industrial tolerance for most machined parts, this accuracy is sufficient to show that a part roughly matches the position and expected geometry of the model. This information allows the machine to monitor for any significant interference or interruptions during and after machining.Item Development of a Scalable Machining Feature Recognition System(University of Waterloo, 2023-12-19) Lenover, Michael; Bedi, Sanjeev; Mann, StephenIn this thesis, various pre-processing and training techniques were applied to improve the performance of a model trained with an existing machining feature recognition approach by Yeo et al. using a smaller dataset that more effectively mimics the complexity of CAD models used in industry. A GUI tool was developed to tag faces in CAD models with the corresponding machining features which would be necessary to resolve those faces. Using the encoding algorithm outlined by Yeo et al., a tool was developed to generate feature vectors from tagged CAD models. Two CAD datasets were compiled. First, a dataset of generic CAD models was filtered from a larger dataset compiled by Koch et al., selecting those models which could be manufactured using a 3-axis CNC machine. Second, a dataset of real-world CAD files used in CNC manufacturing was compiled from models contributed by individuals from the Unviersity of Waterloo, Hurco Inc. and Perfecto Inc. Using the first dataset, three potential improvements to the feature recognition algo- rithm developed by Yeo et al. were explored: the incorporation of dropout to improve model stability and accuracy, the incorporation of ID3 tree pre-classification to reduce training time by reducing the size of the deep learning dataset without impacting classification accuracy, and the incorporation of crossover data generation to improve classification ac- curacy by reducing overfitting due to insufficient training data. It was determined that incorporating dropout improved the stability of the model and improved 5-fold cross val- idation accuracy. Further, it was determined that incorporating a 2-deep ID3 decision tree pre-classification marginally improved classification performance and was effective in reducing the size of deep learning training dataset. Crossover data generation did not improve model performance, and so was rejected. Using the model trained on the generic CAD dataset, and incorporating 10% dropout and a 2-deep ID3 tree, models from the real-world dataset were classified. This classifier was effective in classifying some simple features, but had poor accuracy overall. To improve this accuracy, an incremental learning technique was applied. The generic model was re-trained using samples from the real-world dataset, which improved the classification accuracy of the system.Item Fast and Robust Approach to Find the Gouge-free Tool Position of the Toroidal Cutter for the Bézier Surface in Five Axis Machining(University of Waterloo, 2020-05-15) Singh, Mukhmeet; Bedi, Sanjeev; Mann, StephenOne of the approach used for tool path generation for Bézier surfaces is the Multipoint machining (MPM) approach, in which the toroidal cutter touches the machined surface at two points of contact. Multipoint machining helps in reducing the machining time by providing the tool path data that machines the surface in wider strips positioning the tool in the close proximity to the surface. The tool path generation using MPM is computationally expensive and time consuming, as it involves the solving of non-linear transcendental equations that require numerical methods. Numerical method such as Newton’s method are a time consuming and iterative process, and are not always able to give a solution. In this work, two methods, the ‘Drop, Rotate and Drop (DRD) method’ and the ‘Vertical and Circular Ray Firing (VCRF) method’, are developed, implemented and tested on bi-cubic Bézier surfaces using a Hi-Dyn tilt-rotary simultaneous five axis machining center. These methods follow the Multipoint machining approach. The DRD method limits the use of Newton’s method for convergence to the solution of two unknowns or variables. Whereas, the VCRF eliminates the use of Newton’s method for obtaining the solution and instead uses the implicit equations for firing the rays vertical or circular from the surface towards the toroidal cutter surface. Hence, the methods developed in this work give a fast and robust approach for generating tool path data for the Bézier surfaces.Item Making CNC Machines Smarter(University of Waterloo, 2019-12-16) Kvitnevskiy, Ilarion; Bedi, Sanjeev; Mann, StephenCNC machines are a commonly used manufacturing tool. Over the years, they have become increasingly sophisticated. While there is a lot of research into making the machines more sophisticated, there is little research into making the machines smarter. CNC machines lack any intelligence to make decisions. Making a system fully intelligent is extremely difficult to do in one step. This thesis will focus on small steps that will hopefully lead to an intelligent CNC machine. The thesis first explores using audio data for perceiving the cutting state of the machine. Experienced machinist can listen to the machine and determine how it is cutting and can assess changes for improving the cutting rate or surface finish. Ideally, the machine should be able to determine how it is cutting and use that information to adjust machine parameter for a cutting goal. In this project, a neural network was trained to detect the presence of chatter. Unlike conventional methods, this project involved only doing a Fourier transform of the audio data. The neural network had success in identifying chatter in the audio data in all the cases that were tested. Next the thesis explores incorporating a model of the cutting process and using it to generate its own toolpaths. This method involves using a cutting model that uses 2D pixels for determining the cut and uncut area. Using this model, a tool path is generated by optimizing each step to achieve an optimal cutting goal. Further, constraints are added to the optimization, which improve the toolpath by limiting the turning radius, which makes the path smoother. The result is a toolpath that maintains a consistent cutting force, and smooth turning. The previous project relied on a simplified model of the cutting process. As CNC machines become smarter, they will need to have more accurate models of the process. Part of this would be to have accurate dynamic models of the machine. The last project focuses on building an automated device for capturing such models. This device uses a novel approach compared to traditional tap testing. The devices uses a voice coil for actuation, a load cell for force measurement, and a laser displacement for measuring the vibrations. This allows the tap tester to be able to measure many different tools without manually attaching accelerometers to each tool manually.Item Modelling Microstructure-Property Relationships in Polycrystalline Metals using New Fast Fourier Transform-Based Crystal Plasticity Frameworks(University of Waterloo, 2019-02-04) Nagra, Jaspreet Singh; Inal, Kaan; Bedi, SanjeevThe present thesis develops several new full-field, fast Fourier transform (FFT)-based crystal plasticity modelling tools for microstructure engineering. These tools are used to explore elasto-viscoplastic deformation, localized deformation, 3D grain morphology, microstructure evolution, dynamic recrystallization and their effects on formability of polycrystalline metals with particular attention paid to sheet alloys of aluminum and magnesium. The new FFT-based crystal plasticity models developed in this work overcome several inherent problems present in the well-known crystal plasticity finite element method (CP-FEM) and elasto-viscoplastic fast Fourier transform method (EVP-FFT) in solving representative volume element (RVE)-based problems. The new models have demonstrated significant fidelity in simulating various deformation phenomena in polycrystalline metals and prove to be faster and accurate alternatives for obtaining full-field solutions of micromechanical fields in aluminum and magnesium sheet alloys. In particular to the aluminum alloys, which are currently replacing heavier steel parts in the automotive industry, the sheet aluminum alloys have significantly improved corrosion resistance and strength-to-weight properties in comparison to steel. However, aluminum alloys are still outperformed by steel in terms of formability. To improve the formability of an aluminum sheet, one method is to develop physics-based predictive computational tools, which can accurately and efficiently predict the behavior of aluminum alloys and thus allow designing the microstructure with desired properties. Accordingly, in first part of this thesis, a novel numerical framework for modelling large deformation in aluminum alloys is developed. The developed framework incorporates the rate-dependent crystal plasticity theory into the fast Fourier transform (FFT)-based formulation, and this is named as rate tangent crystal plasticity-based fast Fourier transform (i.e., RTCP-FFT) framework. This framework is used as a predictive tool for obtaining stress-strain response and texture evolution in new strain-paths with minimal calibration for aluminum alloys. The RTCP-FFT framework is benchmarked against an existing FFT-based model at small strains and finite element-based model at large strains, respectively, for the case of an artificial Face Centered Cubic (FCC) polycrystal. The predictive capability as well as the computational efficiency of the developed framework are then demonstrated for aluminum alloy (AA) 5754. In the second part of this thesis, the RTCP-FFT framework, developed earlier, is coupled with the Marciniak and Kuczynski (MK) approach to establish a new full-field framework for generating forming limit diagrams (FLDs) of aluminum sheet alloys, e.g., AA3003 and AA5754. The new coupled framework is able to investigate the complex effects of grain morphology, local deformation, local texture and grain interactions on the predictions of forming limit strains. This study reveals that among the various microstructural features, the grain morphology has the strongest effect on the predicted FLDs for aluminum alloys. Furthermore, this study also suggests that the FLD predictions can be significantly improved if the actual grain structure of the material is properly accounted for in the crystal plasticity models. In addition to aluminum alloys, magnesium alloys are getting significant attention by the automotive industry due to their light weight and high specific strength. However, the automotive industry has not been able to take full advantage of the lightweight characteristic of magnesium alloys because of their poor formability at room temperature. Therefore, to enhance the workability and restore their ductility, the magnesium alloys are formed at elevated temperature. High temperature forming of magnesium alloys is often accompanied by dynamic recrystallization (DRX), which allows the final microstructure, as well as the properties of the material (e.g., initial grain size, initial texture, etc.), to be controlled. Therefore, DRX coupled with a full-field crystal plasticity FLD framework can be used as a tool to design microstructure of a material. Since it would be beneficial to be able to redesign the material properties of magnesium alloys using physics-based computational tools than using physical experiments, this work takes a step ahead towards such an outcome by presenting a new framework that predicts DRX and models its effects on the formability of magnesium alloys. Accordingly, in the third part of this thesis, a new full-field, efficient and mesh-free numerical framework, to model microstructure evolution, dynamic recrystallization (DRX) and formability in hexagonal closed-packed (HCP) metals such as magnesium alloys at warm temperatures, is developed. This coupled framework combines three new FFT-based approaches, namely: (a) crystal plasticity modelling of HCP alloys, (b) DRX model, and (c) MK model. First, a rate tangent-fast Fourier transform-based elasto-viscoplastic crystal plasticity constitutive model for HCP metals (RTCP-FFT-HCP) is developed. Then, it is coupled with a probabilistic cellular automata (CA) approach to model DRX. Furthermore, this new model is coupled with the Marciniak-Kuczynski (M-K) approach to model formability of magnesium alloys at elevated temperatures. The RTCP-FFT-HCP model computes macro stress-strain response, twinning volume fraction, micromechanical fields, texture evolution and local dislocation density. Nucleation of new grains and their subsequent growth is modeled using the cellular automata approach with probabilistic state switching rule. This framework is validated at each level of the coupling for magnesium sheet alloy, AZ31. First, the RTCP-FFT-HCP model is validated by comparing the simulated macro stress-strain responses under uniaxial tension and compression with experimental measurements at room temperature. Furthermore, the texture evolution predicted with the new model is compared with experiments. The predictions show a good agreement with experiments with high degree of accuracy. Next, the forming limit diagrams (FLDs) are simulated at 100 C, 200 C and 300 C, respectively, for AZ31 sheet alloy considering the effects of DRX. The predicted FLDs show very good agreement with the experimental measurements. The study reveals that the DRX strongly affects the deformed grain structure, grain size and texture evolution and also highlights the importance accounting for DRX during FLD simulations at high temperatures.