ME 338 Term Paper Robotic additive Manufacturing Aryan Kumar 210100030 1. Introduction Robotic additive manufacturing is a process of creating an object by adding material layer by layer using different materials, such as plastics, metals, and ceramics. The inspiration for RAM procedures comes from Chuck Hull's invention of the novel technique stereolithography in the 1980s. He developed this technique after getting frustrated with the lengthy production periods of prototyping, which uses UV lasers to build 3D objects, layer by layer. Now it is referred to as “3D printing”. This process was first employed to create conceptual models. It is the main aspect of discussion of design concepts, and for applications involving the process, or for the creation of architectural & anatomical models. 1.1. Why? This process is defined for the production of any object in any shape without the need of tools. In this process, simple and cost-effective construction of complex objects in a single step can take place. Because these procedures require fewer tools and equipment, inventory costs, manufacturing time, material usefulness, and production waste are all less as compared to subtractive manufacturing AM technologies offer us benefits like flexibility of design, fast time for market, environmentally friendly manufacturing practices, and industry-driven assets. AM offers more advantages in terms of money because it doesn't rely on complex geometries to produce goods. from the material deposition system transforms a digital model into a physical component through a series of coordinated and controlled movements. An example of AM in the aerospace industry is GE Aviation's fuel nozzle tip for the ‘Leading-Edge Aviation Propulsion’ engine. Reduction of 25% in mass of fuel nozzle is seen when Am processes are used, this is considered a breakthrough in metal processing. As the industrial robot energy consumption rises, it is becoming very important to consider energy efficiency. Also, when making part placement decisions for sustainable production.Research on energy efficiency in industrial robots for conventional manufacturing has been conducted. for example, for this, a shortest path and energy consumption-based particle swarm optimization technique for spot welding path optimization was proposed. Also, a review paper examines energy-saving techniques for robotic systems from both the hardware and software areas. A robot's energy efficiency is a major factor in determining how useful it is for industry purposes. 1.2. Developments in Additive manufacturing Robotic additive manufacturing has become increasingly popular in recent years, and is now used in a wide range of industries. In the healthcare sector, AM is being used to improve the quality & reduce the cost of patient-centric and regenerative medicine, cardiology, orthopedics, and implants. AM works and an object is created from a digital model by using computer-aided design (CAD) software. A heat source fuel nozzle tip produced by AM (GE aviation, 2018) 2. Methodology 2.1 Robotic Setup CAD packages are realistic simulation environments that are now being used to design robotic cells. This is desirable for rapid and dependable development, configuration, and coding of robotic cells. Additionally, it is necessary to develop and execute software for the remote monitoring and control of cell capabilities. A simple access authorization mechanism (list of allowed IPs and password) is used for robocalls. The robot informs the authorized users/programmers by its command. The user/programmer can obtain the required contour paths from the information in the 3D workpiece file. All the contours are graphical objects, such as lines, arcs, and points, from which the entire path can be identified, just as a CAD engineer uses these commands to design the workpiece. 2.2 Path Generation Designing or creating the path of a tool is a crucial step in additive manufacturing. The path that is generated must be collision-free, accurate, and time-efficient. The design of the tool path is divided into three categories: offsetting the reference surface, slicing the non-planar layers, and generating tool paths for the non-planar layer Researchers developed a new way to plan the path of a tool used to create multi-material assemblies. This toolpath avoids collisions and only requires the tool to fill the assembly. To make products faster using laser-based direct metal deposition, researchers used a robot with 8 arms. This robot has 6 regular arms and 2 extra arms that can tilt. The extra arms can tilt the laser head relative to the part being made, which is held in place by a positioning system. A 6 DOF reconfigurable robot platform 2.3 Energy Consumption We can split Robotic AM energy into ‘AM energy’ and ‘manipulator energy’. Energy needed for printing, including printing energy, preheating, and other misc. is called AM energy. Manipulator energy is used to move the manipulator along the trajectory. We want to build an energy consumption model using losses like frictional, mechanical and electrical losses (wind, core, stator and others). Moreover, using robot kinematics and dynamics to efficiently use energy is better than using losses in controlling applications. We can say that, the energy consumption models based on robot kinematics and dynamics are generally explored. The amount of energy (E) used to travel a certain distance (dλN) during a certain time period (λth time interval) is calculated by dividing the distance by the number of time periods (N) 3. Robotic Additive Manufacturing Processes In 2015, the International Organization for Standardization (ISO) and American Society for Testing Materials (ASTM International) worked together to make a common list of words to describe the basics of different additive manufacturing processes. They divided additive manufacturing into seven categories: 3.1 Binder Jetting Binder jetting uses an ink-jet printing head to drop liquid bonding agent onto a powder material, layer by layer, to join specific areas without heat. It has a robot with 3 arms. through a small opening. It is also known as FDM, fused filament fabrication(FFF), and FLM.. Setup axes range from three to five. 3.5 Powder Bed Fusion Powder bed fusion is a type of 3D printing that uses heat to melt specific parts of a powder bed. Examples include electron beam melting (EBM), selective laser sintering (SLS), and direct metal laser sintering (DMLS). Robotic systems use mirrors to move the heat source across the powder bed in a planned pattern Binder jetting technique 3.2 Material jetting Material jetting is a 3D printing process where tiny drops of material are deposited to form a partly-completed object, which is then hardened. It has a three axes robotic setup. 3.3 Directed Energy Deposition This process uses a focused heat source to melt powder or wire feedstock and join it to a build surface.Examples of this process include using lasers or an electron beam to melt metal and deposit it layer by layer to build a part. Robots with three to seven axes move the laser or electron beam to create the part. The robotic axes range from three to seven. Illustration of Directed energy deposition 3.4 Material Extrusion This is a process where a material is released Powder bed fusion 3.6 Sheet Lamination Sheet lamination is an AM process where thin layers of material are glued together to form an object. One example of this technology is laminated object manufacturing. 3.7 Vat photopolymerization Vat photopolymerization (VP) is a type of additive manufacturing that uses light to harden a liquid resin in a vat. Examples of vat photopolymerization include stereolithography and digital light processing 4. Machine Learning in Robotic AM Finding Input - Output relationship in AM of different processes is very important especially for Wire arc robotic AM. For this, many techniques have been developed.Recently, researchers have started using machine learning (ML) and deep learning (DL) to predict the results of WAAM experiments. This is because the data from these experiments is very non-linear. Deep learning (DL) uses artificial neural networks (ANNs) to make predictions. ANNs are often paired with optimization algorithms to improve their accuracy. Gradient descent-based backpropagation (BP) is the most common optimization algorithm used with ANNs. It works well, but it can get stuck in local minimums. Researchers have tried using other optimization algorithms, such as genetic algorithms (GAs) and particle swarm optimization (PSO), with ANNs. These algorithms work well, but they are slow and require a lot of computing power. Many other optimization algorithms have been developed. One of these algorithms is the gray wolf optimization (GWO) algorithm. GWO has been shown to outperform GAs and PSOs in many cases . 4.1 Experiment design Regression analysis is done to find a relation between input and output parameters. To train the model, we used 1,296 different sets of input values and a nonlinear second order regression equation to create an approximate relationship between the input and output factors. y is the output of a process, and X is a setting that controls the process. b is a number that shows how much y changes when X changes by one unit. e is the error in measuring y. j and i are two different settings that control the process. Xj is the linear effect of setting j, Xj^2 is the nonlinear effect of setting j, and XiXj is the interaction between settings i and j. 4.2 Neural Network Architecture The skills and abilities of the human brain are a result of the billions of interconnected neurons that make up the brain. These mathematical models based on neurological systems are called Artificial Neural Networks (ANNs). They have been widely used for classification issues, predictive jobs, optimisation, and many other activities in a variety of disciplines. Information is fed into the input layer of an artificial neural network. The information then travels to a hidden layer of interconnected neurons, where it is weighted and biased. From there, we get the final output. To train two artificial neural networks (ANNs), we fed the inputs (WFS and TS) to the forward mapping model and got the outputs (BW and BH). We used the opposite parameters for the back mapping model. We started with random values for the weights and biases of the ANN model, between -1 and 1. We used a special function called ReLU in the input and hidden layers of the model. We trained the model on a dataset generated from the experimental data, with the goal of minimizing the error. We then used two different algorithms, gradient descent with momentum and the Grey Wolf optimization algorithm, to compare their performances. 4.3 The gradient descent with momentum Gradient descent is a way to find the best solution to a problem by following a path of decreasing error. The speed of this path is controlled by a setting called the learning rate. Momentum is a method added to the algorithm to ensure that it searches in the right direction without getting lost in the search space while looking for the best solution. The Grey Wolf algorithm is a way to find the best solution to a problem by mimicking the social hierarchy and hunting behavior of gray wolves. The hierarchy is made up of four types of wolves: alpha, beta, delta, and omega. The alpha wolf is the best solution, the beta and delta wolves are the second and third best solutions, and the omega wolves are the rest of the solutions. The algorithm works by imitating how gray wolves find, circle around, and attack their prey 5. Results and Conclusions Robotic additive manufacturing (RAM) is a new way to 3D print things using robots. It is still being developed, but it has the potential to revolutionize the way we build things in the future. RAM can be used to print things much faster and more accurately than traditional 3D printing methods. It can also be used to print things in more complex shapes and sizes. One of the challenges of RAM is creating toolpaths (the paths that the robot follows when printing) that are efficient and collision-free. Researchers are developing new toolpath strategies to address this challenge. RAM is also being used in hybrid manufacturing, where it is combined with other manufacturing processes, such as machining and welding. This allows manufacturers to create complex products that would not be possible with any one process alone . Energy and quality metrics are formulated and a systematic methodology based on single-objective optimization and energy-quality map is introduced. It is used to identify the optimal part placement within the robot’s reach. As industrial robots become more capable and flexible, researchers are increasingly interested in using them for ground and aerial mobility in manufacturing Moreover, Additive manufacturing systems can help products get to market faster and be more customized. However, current AM processes have limitations, such as small product size, slow build rates, and the need for support structures for areas with overhangs. This is driving researchers to develop new AM methods. Two major drivers of change are the three-axis layer-by-layer manufacturing process of conventional AM systems and their limited work envelope. We used two machine learning models to map forward and backward, with three hidden layers of neurons in each model. We trained the models using gradient descent with momentum and GWO algorithms. For forward mapping, the ANN model trained with gradient descent with momentum performed the best, based on its MAPE score on the test dataset. GWO converges much faster than gradient descent. For backward mapping, we found the best number of iterations and hyperparameters by running the model multiple times until we got the desired results. 8. References 1. J. Norberto Pires, Amin S. Azar, Filipe Nogueira, Carlos Ye Zhu and Ricardo Branco, The role of robotics in additive manufacturing: review of the AM processes and introduction of an intelligent system, Volume 49 · Number 2, 2022, 311–331 2. Suyog Ghungrad, Abdullah Mohammed, Azadeh Haghighi. Energy-efficient and quality-aware part placement in robotic additive manufacturing. Journal of Manufacturing Systems 68 (2023) 644–650. 3. J. Norberto Pires, Amin S. Azar. Advances in robotics for additive/hybrid manufacturing: robot control, speech interface and path planning. Industrial Robot: An International Journal Volume 45 · Number 3, 2018 , 311–327. 4. Damir Godec · Joamin Gonzalez-Gutierrez · Axel Nordin · Eujin Pei · Julia Ureña Alcázar Editors. A Guide to Additive Manufacturing. 2022.