Summary
Nobody wants defective steering or brakes on their car, but it’s not easy to find defects in aluminum parts the moment they’re manufactured. Developers working for automotive parts maker Fagor Ederlan were tasked with building a system that could use the available manufacturing data to predict quality issues. Their work was made more challenging by a diverse range of legacy and proprietary equipment, data access issues, sheer data volumes, and network constraints. Using a Microsoft Azure cloud solution and machine learning model, a joint developer team was able to create a prototype solution that identifies defective parts sooner while conserving network bandwidth. Most importantly, it can be applied to the different machines and factories in the organization’s global system.
With this project, Microsoft and Fagor Ederlan demonstrate that sending data to the cloud using secure and reliable services doesn’t have to be overly complex. This allows for data analysis in near-real time and the ability to build machine learning models to extract insights.
The project is based on a molding machine that takes measurements each millisecond. The data captured during the molding process is useful in building machine learning models that indicate when the quality of automotive part not within standards.
Unique technology challenges for the automotive customer
Fagor Ederlan’s machine has control software that gathers all the sensors’ information as CSV files. For each piece, it calculates averages, means, and other statistical values from the data. To minimize impact on the machine, a lightweight Windows Service was developed. Whenever it detected a new CSV file in a folder, the data was sent to IoT Hub.
The remainder of the process occurs on Azure. To develop the service, they took into account all the communications concerns, created a test plan, and tested the service under the most probable situations, such as network outages and other connectivity problems.
To minimize these challenges, the team reduced the size of zip files and deleted files once they’re available on the cloud, providing two machine learning models to gather data.
Automated data gathering
The objective of the solution is early identification of defective pieces in an aluminum molding machine. During the injection process, the machine takes many parameters per millisecond, such as speed, pressure, and injector run. This creates an 800 KB CSV file with all the measurements and another one with averages. The process of filling the die and cooling the piece takes between 60 and 90 seconds. When the piece is finished, there is an X-ray and visual inspection of the pieces to detect defective ones. But the final check comes from the customer, which usually happens one month after the piece is built.
Developers share knowledge and solve company’s challenges
The engagement allowed developers to find a reliable solution for sending data to the cloud and to enhance the security and reliability of the system. The teams used Azure technologies and Power BI to take the following measures:
- Connected Windows Service to IoT Hub to gather and compress CSV documents and other direct sensor information.
- Used a command-line simulator to test the full solution without deploying inside a running machine.
- Took advantage of two Azure Functions: one for decompressing files and the second to extract the characteristics from the curves.
- Used Stream Analytics to join all the data from different files and ask the Azure Machine Learning model
- Retrained an Azure Machine Learning model with new data.
- Deployed the solution multiple times with an Azure Resource Manager template.
- Used a Power BI dashboard to see curves and an Azure Machine Learning–trained model to see data for each piece in real time.
Next steps
Automating file uploads and data preparation has solved critical problems for Fagor Ederlan. The results of the engagement also opens doors to tackle issues that will save costs, time, and enhance quality of the product. Fagor Ederlan wants to continue evolving and improving the scalability of the system with IoT Hub, Stream Analytics, and Azure Functions.
Get the code samples and architecture diagrams from this project on GitHub, get hands-on with Azure IoT Hub labs, and start building an IoT solution with the Azure trial.
