Vertical AI for manufacturing 4.0

Revolutionize your CNC Operations with cutting-edge physics enabled machine learning and optimization tools.

Gears are pivotal in various industrial sectors, such as automotive, energy generation, and defense, with grinding being the primary technology for producing large-scale, high-quality gear components. In the United States, a 6.4% growth in gear demand is expected, reaching $40 billion in sales

However, gear grinding, a subtractive manufacturing process, exhibits high specific energy consumption compared to other machining methods, highlighting the need for holistic optimization for energy efficiency in complex CNC gear grinding operations. The escalating trend in energy costs emphasizes the importance of manufacturing process optimization to reduce expenses and carbon emissions.

Key Points

• High Energy Consumption: Gear grinding has a high specific energy consumption.
• Economic Impact: Energy costs are rising, making optimization crucial
• Environmental Responsibility: One kWh corresponds to 0.433 kg of CO2 emissions

Vertical AI for Manufacturing 4.0

Machine learning and AI are a top priority for many manufacturers trying to improve the efficiency, sustainability, or reliability of their CNC machining processes. However, a major barrier of data-driven solutions in this space is the black-box nature of many solutions increasing the perceived risk of implementation. At the same time, extensive process knowledge has been built over decades trying to understand the physics of manufacturing processes, in particular behind material removal. Merging physics-based and data-driven modeling approaches have the potential to alleviate these concerns and align with the conditions and requirements on the manufacturing shop floor. However, knowledge of how to approach developing hybrid models for CNC machining processes presents a significant challenge for manufacturers. This practical guide to hybrid modeling in manufacturing aims to address this challenge by providing practitioners with a hands-on tool to develop solutions for their own CNC machining processes. The insights are based on experience from a recently completed collaborative project optimizing the CNC grinding processes of large gears, that yielded a reduction in energy consumption by 37% and a reduced processing time by 41% through the application of hybrid modeling in a live production environment. Hence, this guide is purposely written with practitioners in mind.

Hybrid Model Composition

The mode of combining the physics model and data-driven models is quite crucial as it dictates the effectiveness of the hybrid model. Typically, there are three different strategies to combine physics with the pure data-driven model (Figure 6). First, physics can be incorporated as an additional constraint in the loss function of the data-driven model. Nonetheless, complex constraints result in difficulties in the training procedure. Second, the data-driven model can replace the computationally expensive component in the physics model. However, the physics models are commonly complex and highly coupled, which makes these methods challenging to apply. Third, the output of the physics model is directly combined with the data-driven model as an extra input to the data-driven model, which is also called the combination approach. Compared with the first two kinds of interactions, the combination approach is the most straightforward and requires minimal changes to how the data-driven model is trained. Therefore, the combination approach is selected to combine the physics and DNN models to enhance the overall predictive capabilities.

The Savings

Key Points

• Optimization Factors: Time, materials, tool life, and process parameters
• Industry 4.0: Real-time monitoring and control for cost-effectiveness and sustainability
• Significant Reductions: Achieved a 37% reduction in energy and a 41% reduction in time