Modern automotive plants are generating incredible amounts of data from machinery, sensors, production tools, inspection equipment, and tracking systems. The challenge is distilling that data into actionable insights for production or maintenance personnel.
That is where AI and machine learning can come in. The key is to look at it as an enhancement to existing processes. AI models can be connected to existing factory data and systems in order to recognize patterns, detect anomalies, and aid in decision making.
- Predictive Maintenance
Production equipment can generate data from temperatures, vibrations, pressure, current, and more. A Machine Learning model can learn characteristics of normal equipment function and determine when anomalous data points occur.
For instance, a machine learning model could be used to detect when a machine is displaying patterns that require maintenance. An example of this might be:
Machine → Sensor → Data Gathering → ML Model → Anomaly Detection → Maintenance Procedure
The model does not necessarily need to act as a replacement for maintenance workers. It can instead be used to give earlier warnings or more information to maintenance personnel.
Another application of AI in this space is computer vision. This field of study is a viable application for automotive production facilities.
Computer vision works by using a camera to capture images of a target, and then using a trained model to evaluate the image and look for specific features. A simplified version of an inspection system might be:
Industrial Camera → Preprocessing → Vision Model → Inspection Results → Integration With Production Workflow
Human inspection can be used for either edge cases that the model fails on, or as a secondary review to ensure quality.
- Production Optimization
Production lines are full of data about cycle times, machine downtime, production schedules, material location and movement, and more. By analyzing this data, companies can look for trends or insights that improve productivity.
For instance, some companies utilize analytics platforms to investigate issues with production downtime, or to look for factors that contribute to delays in production. The critical factor is that the question being asked is focused, and the analysis provides value to a specific process.
- Asset and Material Tracking
Large automotive plants often have thousands of different assets or materials that need to be tracked at any given time. Different tracking solutions can be used to identify the location or movement of inventory within a plant. Some vendors offer RFID, BLE, and UWB tracking systems that can provide valuable data when combined with other tools. Using this information in a broader analytics platform can provide valuable data for facilities looking to optimize material flow.
Some potential questions to ask themselves about material flow might be:
Where is this asset/material?
How long has it been there?
Is it moving to the next location on time?
Are there certain points that cause material delays?
- Integrating AI Into a Larger Industrial System
When it comes to industrial AI applications one of the most important considerations is integration. Many plants have legacy systems such as MES, ERP, SCADA, PLCs, IoT systems, and databases that are used for day to day operations. An isolated ML model that provides valuable insight but lacks integration with these systems is significantly less valuable than one that can plug into a larger system. A system architecture that facilitates that might look something like:
Factory Equipment → Industrial Data → Processing/Edge Layer → AI/ML Model → Insights → Existing Manufacturing Workflow
This way, the AI model has value beyond an academic or isolated research model.
If you want to see a practical application of AI, Industrial IoT, tracking systems, and automotive manufacturing technology working together, you can view an example application at OEMNex AI(https://oemnexai.com/).
What are Practical Considerations Before Adopting AI?
Before adopting AI, some practical considerations should be made by the manufacturing facilities that wish to utilize this technology.
Some of the critical questions to ask before developing a model include:
What problem are we trying to solve?
Do we have high quality data?
How are we going to deal with poor or faulty data?
How much does it matter if the model is wrong?
Will this run on the Edge or in the Cloud?
How can we integrate this with our existing systems?
How do we handle uncertain data?
How do we monitor the model after deployment?
These questions are important to ask because model creation is only the beginning of the larger process. A comprehensive implementation requires knowledge in data science, computer vision, industrial connectivity, hardware and software development, and more.
Final Notes
AI applications are being used in diverse fields, but it is essential to remember that most applications come with serious costs, limitations, and risks. For every successful AI application, there are countless failed attempts, abandoned projects, or white elephant systems that never see the light of day.
Top comments (0)