The Digital Chisel – Forging a Collision-Free Future in Computer-Aided Manufacturing
In the following article, Yavuz Murtezaoglu, Founder & Managing Director of ModuleWorks, traces the evolution of simulation technology in CAM over the past 30 years, highlighting its progression from basic backplotting techniques to advanced solutions like 5-axis machining, collision detection, real-time collision avoidance, material removal simulation, GPU acceleration, and efforts to establish universal standards and cross-platform compatibility.
In the world of computer-aided manufacturing (CAM), simulation technology has evolved from a luxury to what many consider an essential component. As we explore the landscape of CAM software today, three questions consistently emerge: Is simulation of tool paths in CAM truly a must have or merely a nice-to-have feature? What capabilities should we expect from simulation in today’s modern CAM software and powerful hardware? And perhaps most importantly, what developments can we anticipate in the coming years?
I would like to take you on a journey through my last 30 years in this field, sharing insights and findings from my experience. This personal perspective will trace the evolution of simulation technology from its early rudimentary forms to today’s sophisticated solutions, providing a comprehensive understanding of where simulation technology stands in modern CAM.
I would like to take you on a journey through my last 30 years in this fi eld, sharing insights and findings from my experience. I hope you will find it both interesting and informative, providing a comprehensive understanding of where simulation technology stands today in CAM.
The Pioneering Phase: Breaking New Ground
In 1997, 3D CAM was still very new. The primary challenge was educating manufacturing companies about the necessity of CAM software, post processors, and data import capabilities from different CAM systems. Five-axis machining was an even smaller niche, and professionals used the term ‘True 5-axis machining’ to describe the continuous motion of all five axes simultaneously. The majority of customers were performing what we called ‘indexed’ or ‘3+2 axis machining’, where the machining direction remained fixed for each tool path operation.
I was fortunate to be in Germany, where machine tool vendors were increasingly introducing 5-axis CNC machines to the market, and control makers like Siemens and Heidenhain were enhancing their capabilities to support these advanced machines.
During this period, I was developing 5-axis tool path algorithms to overcome CAM software limitations, enabling customers to machine complex shapes across various industries. However, determining whether a tool path was safe presented significant challenges. The integrated simulation in CAM software utilized backplot technology to display the tooltip as a series of lines— effective for 2D or 3-axis machining but inadequate for 5-axis machining, where the tool tip could maintain its position on the line while the tool itself tilted.
The material removal simulation integrated into these systems effectively demonstrated how parts were created from stock material but failed to identify potential collisions when all machine components were in motion. Dedicated standalone simulation software was available, but it required separate licensing, installation, and considerable patience to evaluate each program.
Beyond Backplot:
The Machine Kinematics Revolution
We recognized that the most significant risk in 5-axis machining was machine crashes. To address this, we began simulating complete machine kinematics. Given the complexity of machine geometry, we focused on simplifying it, concentrating on critical components such as the table, work holding, and spindle while disregarding the housing and other less critical details.
We discovered that OpenGL, well-developed for gaming, could be leveraged to create fluid animations of machine movements for any 5-axis tool path. The challenge was converting CAM software tool paths to machine motion. The solution lay in the post processor: its role is to transform the tool path for the part to machine kinematics, mathematically converting the tool axis vector in the workpiece coordinate system into rotary axis angle values for the specific machine.
Since 5-axis post processors were rare, we developed this capability in-house. Our post processor developers provided the kinematic solution, which we connected to an OpenGL engine. The result was a remarkably fast capability for simulating tool paths for any 5-axis CNC machine. By the early 2000s, we showcased this technology at a major exhibition on a large screen, creating what seemed like magic: complex 5-axis programs simulated rapidly for any kinematics.
While fast, this system initially didn’t report collisions. Nevertheless, it proved valuable. We termed it ‘Visual collision checking’, similar to tool path backplotting where users manually inspect the tool path. Operators could run the simulation, rotate the view with the mouse, and visually determine if collisions would occur. Given the comparative slowness of computers at that time, this represented a pragmatic solution, offering instant simulation without waiting time. Users could navigate between the program’s start and end points simply by moving a slider bar with the mouse, visualizing all motions.
We decided that our 5-axis tool path should be used in conjunction with this simulation and bundled them together to prevent machine damage due to inadequate simulation capabilities. While some customers purchased additional standalone simulation packages, most found our fully integrated simulation extremely helpful since it utilized the same kinematic solver as the post processor they employed to operate the machine.
From Visual to Virtual: Perfecting Collision Detection
As computer hardware became increasingly powerful, our ambitions grew: why not implement actual collision checking instead of merely visual collision checking? Encouraged by OpenGL’s success in gaming and how it facilitated our solution development, we explored further gaming technology and collaborated with an expert who had developed collision detection engines for the gaming industry. After significant investment and adaptation to meet industrial needs, we achieved full collision checking, also known as ‘clash detection’.
Accustomed to instant simulation capability, we aimed to maintain similar speed levels even with collision checking. This presented challenges, but we discovered that by reducing the number of triangles in the triangle mesh model, we could achieve satisfactory speeds. Additionally, we defi ned specific collision pairs rather than checking everything against everything. For instance, since a machine’s table would never crash against its housing, we eliminated that check. The outcome was extremely satisfactory and quickly became integral to our solution.
Collision Avoidance System (CAS) of ModuleWorks
At this stage, however, we were only addressing machine collisions and gouges of the tool and spindle against the target workpiece geometry. While we could guarantee that the target part wouldn’t be damaged by the tool and no machine collisions would occur, we lacked material removal simulation, meaning unexpected cuts into unmachined stock went undetected, such as a rapid motion crossing the stock. This required material removal simulation, so we directed users to run the integrated material removal simulation engine within their CAM software, typically licensed from specialist companies providing such technology.
We initially thought this would conclude our journey, as our primary objective was to help users run our 5-axis tool paths safely, and we had developed a working solution that satisfied our customers.
The Digital Chisel:
Revolutionizing Material Removal Simulation
In 2005, at an academic conference, I met Dr. Stautner from Dortmund University, who had completed his PhD on material removal simulation technology and subsequently joined our team. We were receiving numerous customer requests for fully integrated material removal simulation within the machine simulator.
Upon examining existing market technology, we found that most CAM software utilized a technology based on mesh Booleans, where tool motion is described as a mesh subtracted from the stock mesh. While initially quick, the process slowed dramatically as more Booleans generated increasingly more triangles. We adopted the discrete model originating from Dortmund University but recognized the substantial work required to ensure accuracy. Though fast, making the results visually appealing and supporting technologies like turning and wire cutting presented significant challenges.
Since then, we’ve invested approximately 200 person years in development, including quality assurance and product management, to bring the technology to its current level—an achievement we’re extremely proud of. As developers, we tend to seek elegant, simple solutions to complex problems, but this particular challenge demanded extensive effort since every detail is critical to avoid misleading users. The solution must be absolutely safe, fast, user-friendly, and integrable.
Preventive Intelligence: Real-time Collision Avoidance
By 2015, our journey continued despite our satisfaction with our simulation technology, which had been integrated into numerous CAM software systems for material removal and machine simulation, with our market share growing steadily. A new challenge emerged: how to run the simulation engine on an industrial PC adjacent to CNC control to prevent machine collisions in real time.
This exciting challenge came with the advantage of receiving information about future machine movements with a one-second look-ahead, providing data of the ‘future’ one second in advance. Our task was to calculate collisions and material removal and stop the machine before any issues occurred. This necessitated significant optimization of our calculation engine and required heroic efforts from our team to deliver timely solutions to partners. We named this technology CAS (Collision Avoidance System) and believe it could eliminate all machine tool crashes if widely implemented. With the rapid advancement of chip technology, accelerated by developments in AI, we anticipate that within a few years, even basic chips in CNC controls will have sufficient performance, eliminating the need for additional industrial PCs.
Breaking Barriers:
Universal Standards and GPU Acceleration
While CNC machines can avoid collisions using CAS, this requires proper definition of tool, holder, work holding, and stock geometry for each job to enable CAS functionality. Many CNC machines don’t require such data to cut parts, but it’s essential for collision avoidance. Upon investigation, we determined that all this data exists within CAM software but lacked a standard format for export to CNC machines – typically, only the NC program was transmitted.
We initiated the development of MDES (Manufacturing Data Exchange Specification) to enable the export of job setup data from CAM software to CNC machines running CAS. Working with approximately 90% of major global CAM vendors and most CNC control makers and machine tool vendors, we secured substantial support from key industry players. The adoption of this workflow is progressing impressively. To accelerate adoption, we’ve made this specification freely available as an open standard to prevent the proliferation of competing proprietary standards.
Representation of a GPU
The story doesn’t end there. NVIDIA’s success with AI has driven the development of powerful GPUs previously primarily used for gaming. With stable development environments now available for industrial GPU applications like simulation, we’ve ported the most computationally intensive parts of our simulation engine to GPU. While maintaining all features developed over the past 20 years, we now benefit from GPU power. The benchmark results on a mid-range GPU are fascinating: many tool paths with 1-3 million lines of NC code complete simulation at the highest resolution in under 10 seconds. I confidently assert that material removal simulation should never take more than 10 seconds, regardless of tool path size. We’ve achieved this without compromising quality or taking shortcuts to increase speed, maintaining the extensive detailed work we’ve invested (the 200 person years of effort since 2005). This approach facilitates straightforward retrofitting for all CAM software companies using our solution, they needn’t change anything for their users to enjoy this dramatically improved performance.
While one team focused on GPU simulation, another team worked with numerous developers for years on triangulating our discrete model. We discovered that our discrete model offered many advantages, including linear speed increases with more moves, numerical stability, and perfect memory usage control. However, the triangulation was based on discretization. We found a method to generate perfect triangulation based on customer tolerance requirements without generating excessive or insufficient triangles while preserving features like holes, sharp edges, and fillets. This significantly improved simulation speed in ‘play mode’ and substantially accelerated the simulation engine used for updating stock models between consecutive tool path operations such as roughing and rest-roughing, as it generated better and fewer triangles. Improving tool path calculation quality and speed represents a major benefit for all CAM users. We realized that reliable, accurate, and fast (thanks to GPU) in-process stock updates in CAM systems could deliver significant time savings.
It’s exciting to see how our simulation technology contributes to productivity gains for CAM users worldwide. As our CAM partners integrate and release these capabilities, users will soon access this exciting technological suite.
The Future
Are we slowing down? Not at all. Since our solutions are being integrated into existing hardware and software platforms, we’re busy ensuring all our solutions work on ARM CPU architecture, Linux, macOS, and various real-time operating systems. Supporting numerous features across multiple platforms and operating systems necessitates extensive work on automated performance and regression testing. Approximately 30 people work in Quality Assurance, and hundreds of CPUs and GPUs test our code after every single code change submitted to our codebase. While developing complex algorithms is challenging, creating industry-grade complex algorithms is significantly more demanding. This explains why we now employ almost 400 people, most in development, with no time to decelerate.
ModuleWorks is at the forefront of digital manufacturing software, playing a key role in enabling the efficient production of increasingly complex parts in an environmentally sustainable way. As a strategic partner to leading CAD/CAM vendors, CNC control makers, machine tool builders and cutting tool manufacturers, ModuleWorks develops software that powers solutions throughout the manufacturing industry.
For more information: www.moduleworks.com
