Optimizing wheel dressing on CNC grinding machines

Predictive maintenance uses data-driven strategies to anticipate and prevent equipment failures. It involves continuous monitoring of equipment using sensors and advanced analytics to enhance reliability. By predicting when failures are likely to happen, companies can implement more effective maintenance strategies, improving equipment longevity. In this blog, we explore how predictive maintenance is transforming the optimization and efficiency of wheel dressing in CNC grinding machines with a collaboration between HAQM Web Services, Inc. and Petrus Technologies.

AWS offers a comprehensive suite of services that are central to this project. AWS IoT Core provides seamless device connectivity, while HAQM SageMaker is crucial in building and deploying predictive machine learning (ML) models. Petrus Technologies, with its specialization in Industrial IoT (IIoT) and on-premises deployments, ensures seamless integration between cloud-based services and local systems. Their expertise is vital in implementing the predictive maintenance system directly within the operational environment.

Business problem

A leading automobile parts manufacturer faced challenges due to excessive grinding wheel wear. This wear results in frequent wheel dressing and unexpected downtime, disrupting production and affecting product quality. Wheel dressing involves stopping production to remove dull or worn abrasive grains from the surface of a grinding wheel, allowing fresh, sharp grains to take their place. Dressing is necessary because over time, the wheel becomes clogged or dulled with metal particles and loses effectiveness. These issues lead to increased operational costs and reduced efficiency.

This project addresses the following challenges:

·       Reducing Downtime: Optimizes wheel dressing cycles to decrease interruptions in production.

·       Extending Grinding Wheel Life: Enhances the longevity of grinding wheels through better maintenance scheduling.

·      Lowering Maintenance Costs: Reduces the expenses associated with frequent grinding wheel replacements and unexpected maintenance.

Solution

The Petrus Technologies solution uses a sophisticated predictive maintenance model to manage and optimize wheel dressing cycles in CNC grinding machines. This approach relies on advanced ML algorithms to predict the Remaining Useful Life (RUL) of grinding wheels, enabling better, proactive maintenance decisions. Here’s how the solution is structured:

Machine learning model: The ML model uses highly detailed data from FANUC CNC machines, Variable Frequency Drives (VFDs), and quality comparators to create a regression-based ML model using critical parameters like wheel speed, feed rate, and grinding force. This model predicts the RUL of grinding wheels, which is essential for monitoring their current state and performance.

Deployment at the edge: The trained model then deploys via HAQM SageMaker Edge Manager to enable real-time, local processing on edge devices. This setup reduces latency, ensuring that predictions are immediately actionable.

Real-time predictions and alerts: As the model continuously analyzes incoming data, it predicts the necessity for wheel dressing. When the RUL prediction falls below a certain threshold, the system generates automatic alerts to the machine operators through the Human-Machine Interface (HMI). This process helps in scheduling maintenance to prevent failures or quality degradation before machining problems occur.

Dashboard for performance monitoring: A dashboard hosted on HAQM Managed Grafana provides live insights into key performance indicators (KPIs) like wheel life, operational efficiency, and maintenance timelines. This tool helps engineers and supervisors support faster, more informed decision-making.

Feedback loop for model improvement: The architecture includes a feedback loop where operational data and maintenance outcomes are continuously uploaded to AWS. This data refines and improves the machine learning model, maintaining its accuracy as operating conditions evolve.

Through this approach, the solution not only tackles the immediate issues of wheel wear and unexpected downtime but also enhances overall operational efficiency and product quality. The result is a manufacturing process that is more efficient and cost-effective.

System Architecture

Figure 1: Architecture Diagram

The system is broken down into several key components:

OT Network Setup:

• OT and IT Integration: Data is collected from FANUC CNC machines, Variable Frequency Drives (VFDs), and quality comparators and is ingested into PetrusConnect via southbound connectors running on a Petrus Gateway. The PetrusConnect platform facilitates seamless data acquisition and pre-processing within the operational environment.
• Edge Inference: An Open Neural Network Exchange (ONNX) model deployed via HAQM SageMaker Edge Manager processes the data to predict the Remaining Useful Life (RUL) of the grinding wheels. This inference is carried out directly on a Petrus Gateway, ensuring real-time predictions and low latency.

AWS Cloud Integration:

• Data Handling: Data processed within the OT network is sent to AWS through HAQM Data Firehose, which ingests the raw event data into HAQM Simple Storage Service (S3).
• Data Analysis and Storage: AWS Glue organizes the data stored in HAQM S3, enabling structured analysis and providing a foundation for long-term data retention and retrieval.
• Machine Learning: Updates and retraining of the machine learning models are managed within HAQM SageMaker, ensuring that the models continuously evolve and improve based on incoming data.
• Dashboard: HAQM Managed Grafana offers real-time visualization of KPIs, which are essential for monitoring the system’s performance and enabling data-driven decision-making.

Dashboard Insights and KPIs:

Figure 2: Process values view

This HAQM Managed Grafana dashboard process values view shown in Figure 2 includes various KPIs that are important for monitoring grinding wheel wear over time. These KPIs include the minimum, average, and maximum grinding wheel width, which helps track wear and tear. The charts at the top of the display show offset measurements, offering insights into the gradual changes during the grinding process.

A bar graph on the display shows the actual width of the workpiece after grinding. This allows operators to ensure that the grinding process remains within specified tolerances. Additionally, the chart at the bottom of the process values view shows the trend of the current grinding wheel width, helping predict when wheel dressing will be necessary. This prediction ensures that maintenance activities are scheduled proactively, preventing unexpected downtime.

Figure 3: Analytics view

The Analytics view dashboard shown in Figure 3 provides a broader perspective on overall system performance. It includes a cycle count tracker and the Remaining Useful Life (RUL) of the grinding wheels. The cycle count graph helps operators monitor completed cycles, which directly relates to grinding wheel wear.

The RUL metric is valuable as it predicts when wheel dressing will be needed, based on historical data and current operating conditions. Additional metrics, such as motor voltage and frequency, are visualized to offer deeper insights into the machine’s operational health. A quality measurement heatmap provides a detailed view of output quality, ensuring that any deviations are promptly addressed.

These dashboards, powered by AWS services, enable real-time monitoring and predictive maintenance. This capability allows teams to make data-driven decisions, optimize grinding processes, and reduce operational costs across multiple assets.

Conclusion

Implementing predictive maintenance for CNC grinding machines is not just forward-thinking—it is essential in today’s competitive manufacturing environment. Petrus Technologies, together with AWS, helps companies significantly reduce downtime, extend the life of critical components like grinding wheels, and lower overall maintenance costs.

This solution empowers manufacturers to transition from reactive to proactive maintenance, ensuring production lines remain operational and efficient. The system’s ability to evolve with changing conditions through continuous feedback and model updates supports long-term reliability and performance.

As the manufacturing industry increasingly embraces Industry 4.0, integrating predictive maintenance systems like the one outlined here will become a cornerstone of sustainable and profitable operations. By adopting this technology, manufacturers will enhance operational efficiency.

Special thanks to Mahendra Prakash for his help creating this blog. Mahendra is an IIoT and ML Engineer at Petrus Technologies Pvt. Ltd., India. He specializes in building edge runtime systems that integrate with CNC machines, PLCs, and sensors. His work includes data collection, transformation, applying machine learning models, storing processed data, and building unified namespaces (UNS) using MQTT for seamless industrial data management.