by Cassandra Balentine
Artificial intelligence (AI) is a driving force of innovation in nearly every industry segment. When it comes to manufacturing, three-dimensional (3D) printers use AI to improve the experience, efficiency, and scalability of the process both through proprietary and third-party solutions.
Above: 1000 Kelvin AMAIZE uses AI to simulate the metal AM process to increase the efficiency of engineers and avoid build failures. This image demonstrates the actual build process of metal using an EOS printer.
AI in 3D/AM
The implementation of AI into 3D printing and additive manufacturing (AM) systems opens the door to new possibilities by automating and optimizing complex workflows, enabling scalability, and enhancing production quality and efficiency, according to Arvind Rangarajan, global head, software and data, HP Personalization & 3D Printing.
Eefje Verhoelst, R&D manager, AI and data research, Materialise, agrees, noting that AI is already utilized in 3D print/AM environments as a tool to improve operational excellence and increase efficiency and scalability through workflow automation.
Automation within the workflow helps to improve efficiency before the manufacturing process begins, reducing time spent on tasks, such as printability assessment, scheduling, build preparation, nesting, build orientation, and price setting. “Collecting machine data and learning from this data through AI can help in predictive maintenance and root-cause analysis of scrap rate. Generative AI can help in decreasing the learning curve for operators,” explains Verhoelst.
The design stage is another area that presents opportunity. AI is used to enhance processes with algorithms proving to redefine digital models and produce enhanced product designs, shares Rangarajan.
Additionally, Rangarajan sees AI being integrated into maintenance processes to conduct real-time monitoring and predictive maintenance as well as identifying potential issues before the overall production system is impacted. “When integrated with cloud computing, data flows more seamlessly across the manufacturing process, enhancing connectivity along the production line from design conception to post production. Ultimately, we are just scratching the surface with how AI can be implemented in AM.”
In terms of quality and process control, Verhoelst points out that AI provides error detection through pattern recognition, which can identify defects within a build that are difficult for humans to detect.
Fabian Alefeld, senior manager, Additive Minds Consulting, EOS, mentions one tool, 1000 Kelvin GmbH’s AMAIZE, which uses AI to simulate the metal AM process to increase the efficiency of engineers and avoid build failures.
Alefeld feels that AI is just starting to find its way into AM. “From my perspective, AI has proven its first use cases on the largest industries and occupations, and where large language models have great use cases e.g., software engineering. As GPUs become more affordable/adopted and open source models become more applicable to manufacturing, we’ll see a stronger adoption,” he predicts.
Benefits of AI
There are many benefits and potential applications for AI in 3D manufacturing.
One of the most significant benefits of AI in AM is the endless possibilities of learning from collected data. “AI can be combined with domain knowledge but can also be used as such when domain knowledge is lacking. In that way, it has the potential to give new insights for process improvements that would not be achievable without AI. Those AI-based learnings improve product quality, process monitoring, process control, consistency, and efficiency; and therefore leads to cost reduction and time savings,” explains Verhoelst.
At a high level, Rangarajan says AI addresses the scalability and adoption challenges faced by the 3D printing industry. “From a design perspective, AI is both enhancing and simplifying the design process, empowering product designers and engineers to more easily design for AM by enabling faster simulation of product performance. This means that we can turn a group of 50,000 expert additive designers into millions of printable 3D design creators, lowering the barrier to entry for a traditionally technical and niche skillset.”
Rangarajan also sees an impact on total cost of ownership as AI-enabled predictive maintenance, real-time system monitoring, and improved process control reduce errors and variation in printed products while flagging detected issues before they impact a production run, saving time, money, and increasing yield. “3D printing and AI will continue to advance and evolve in lockstep, uncovering new ways to advance both standard and AM processes in this new era of digitized production.”
Recognition of parts through AI helps in build preparation, supporting users and decreasing the learning curve when starting with AM. “Ultimately, these technologies enable greater efficiency and scalability within the AM workflow by providing insights on process improvements and automating digital processes such as the digital planning process of a printed medical implant or guide and digitally recognizing parts after building to assign correct post-processing classes to the parts,” says Verhoelst.
There is significant potential of AI automation in 3D manufacturing that allows for scaling. “One example of this benefit is the analysis of images during the build process to predict part defects. AM parts have hundreds of printed layers, and manually evaluating these in quality control is time consuming and prone to human error. By leveraging AI to automate and monitor the printing process, users can avoid failed builds and hidden flaws in parts and optimize processes to avoid waste in materials and machine time,” shares Verhoelst.
“AM is typically a very complex process, melting thin layers of powder—between 30 to 200 microns—of metal or polymer powder with a laser, layer by layer by layer. A lot can happen over the course of thousands of layers, so the ability to simulate and predict anomalies is very interesting. Also, there are a lot of parameters that engineers have the ability to change in order to achieve a certain outcome. For example, laser power, speed, layer thickness, and gas flow. All of these parameters have various inter-depended outcomes leaving an engineer with endless opportunities. A prime use case for a digital assistant,” suggests Alefeld.
AI Challenges
With any emerging technology there are challenges to contend with.
“The main challenge today is that AM is still a comparable small industry. The required investment into AI tools is high, whereas the return on investment in a smaller industry like ours can take longer. Hence, we have to adopt tools from and partner with other industries, such as conventional manufacturing and find common use cases to push AI together. In an industry where it’s hard to find a workforce, everyone benefits from efficiency tools,” notes Alefeld.
Capturing, processing, and consuming the data needed to inform AI solutions can be a challenge to implementation. “AI requires investment in sensors and internet of things solutions to capture high-quality data that informs effective AI solutions, and also requires measures to ensure the security and privacy of data used,” says Verhoelst. “Building several AI data products based on the collected data typically also imposes the need for some sort of data platform to streamline the data pipelines, ensure data quality, enforce testing and validation, and create version control of the solutions made.”
Once systems are in place to collect data, Verhoelst says the amount can become excessive. “Choices in data collection should still be guided by domain knowledge. In addition, the AI solution itself may predict the outcome, but when wanting to, for example reduce scrap rate, root cause analysis is an essential, challenging part to create impact.”
Rangarajan points out that just as it’s crucial to evaluate the best use cases and applications of 3D printing to ensure the technology suits a particular workflow or production environment, the same careful consideration is needed for AI. “While AI has vast potential to enhance AM, not every application will lead to optimization. For example, the application of AI to physics powered simulations are still in the early stages of development. There are IP issues associated with design generation and there are data privacy requirements and costs associated with collecting, managing, and leveraging customer data for building AI models.”
Rangarajan feels that it is essential to strategically assess where AI can truly add value rather than applying it indiscriminately. “This thoughtful approach requires investment in both time and money to upskill workforces and adapt to a new normal that incorporates greater AI usage and builds trust in solutions.”
“Companies need to look beyond the hype around newer technologies like Generative AI to identify where the technology can truly improve upon existing practices, and where it has been tested and proven in a manufacturing environment,” adds Verhoelst.
To create AI solutions companies require specific skills and competences in the employee profiles typically not yet present in a manufacturing environment, such as experience in software development. “Knowing programming languages like Python, familiarity with practices such as DevOps and MLOps, general data science experience, and preferably cloud expertise,” adds Verhoelst.
AI Functions for AM
AI functions are in their early stages, but are being utilized in 3D/AM today.
EOS is working on various AI projects that are not public yet, “but we believe that AI can be very beneficial for our AM users,” admits Alefeld.
Over the past few years, HP has worked to introduce AI-powered offerings focused on optimizing production and bringing new application designs to life. In Spring 2023, HP announced its 3D Digital Sintering and 3D Process Development software, which uses AI and physics-based modeling to detect, report, and compensate for variability such as shrinkage or deformation of HP Metal Jet-produced parts during the sintering process.
More recently, HP has explored the use of AI in partnership with Shutterstock and its 3D AI generator, which allows designers to rapidly iterate concepts and develop digital assets. Using HP technology, these digital assets are converted to 3D printable models that are then fed into HP 3D printers to manufacture real, physical prototypes to be used to inspire new product designs.
Verhoelst says Materialise leverages AI for recognition and categorization tasks in the AM process. It primarily uses AI in applications to automate image segmentation and classification, both for medical images and build images; to automate part recognition and categorization for part traceability; to facilitate build preparation; and in automating quality control processes, such as error and defect classification. “In these tasks, we find that AI can provide higher accuracy and lower user involvement than traditional algorithms for more consistent and repeatable quality control in the printing process.”
Materialise’s software for 3D printer manufacturers includes the company’s AI-based Quality and Process Control (QPC) Layer Analysis offering. “This module uses AI to examine images of individual printed layers that are captured by cameras on 3D printers. In seconds, the tool scans the layer, identifies errors, and visualizes them to give users a summary of defects in the build. This acts as an assistant to users, enabling them to identify errors that are not easily identifiable to the naked eye and to conduct root cause analysis to determine the impact of errors. This helps users more quickly determine if part quality meets standards and requirements in quality control,” explains Verhoelst.
Virtual Assistants
One of the primary uses for AI in 3D/AM settings is as a virtual assistant.
For example, 3DCeram developed CERIA for its CERAMAKER range of printers. CERIA is deployed in several modules, the first of which, CERIA Set, is a 3D printing assistant designed to check files, compose tanks, and produce optimized printing parameters.
Verhoelst adds that AI currently acts as a virtual assistant in AM quality control by identifying issues and errors within builds. “In the long term, there is potential for AI to assist in optimizing settings and processes for printing, based on previous builds and settings.”
Alefeld also sees this as a key use case for AI in 3D/AM. “AM requires some time to get to excellence as an operator or engineer. Leveraging digital assistants that can guide in the form of an ‘experienced advisor’ would allow us and our users to bring new applications to market faster, cheaper, and with newer talent.”
EOS partners with companies such as 1000 Kelvin, but also has a few projects in its pipeline. “I believe that AI will touch all process steps in the AM value chain, from design to data preparation to post processing. Some of the tools will be enhanced with AI by EOS, some will be provided by the many partners we work with,” notes Alefeld.
“Currently, we see AI virtual assistant solutions playing a significant role during the planning phases, particularly when integrated with third-party design or build preparation software. While these solutions will eventually be incorporated directly into the printers, the factory environment presents both opportunities and challenges. Opportunities include sensor integration, but challenges such as limited compute resources and, in many cases, lack of cloud connectivity, must be addressed,” says Rangarajan.
HP is evaluating virtual assistants that can help its customers to choose the best parameter sets to complement the HP 3D Printing Digital Production Suite and provide design recommendations.
AI in Manufacturing
AI is expected to play an increasing role in AM. Today, it is utilized in the form of virtual assistant technology to help automate and optimize complex 3D workflows.
Nov2024, Industrial Print Magazine
