In the Manufacturing industry (and any industry for that matter), perfect functionalities, top-notch features and highly qualitative aesthetic delivery are essential for remaining competitive in the market. Undoubtedly, stopping errors before they start using intelligent automation is key to aligning products with users’ expectations and to proving your value as a manufacturer. This process of prediction and prevention in due time relies heavily on the precision and swiftness with which sophisticated predictive tools and models can be built and monitored. And it all comes down to one main character: data.
Given the context of tremendous amounts of data being available nowadays, data has unequivocally become the new oil. But data has no value if it does not become information. And this process of transformation involves data processing, analysis and information-driven actions, with strong involvement of the academic world, and following closely the industries' breakthroughs.
AI Services in Accesa
It goes without saying that AI continues to gain new territory as it brings efficiency through automation, predictions, improved compliance and acts, in the end, as a vital performance force. As to our AI experience, it began in 2018, with converting a fully manual work of processing thousands of images, in an automated project that used a neural network doing the work almost entirely by itself. Think about the anonymization done in Google Maps, which consists of obfuscating people's faces and car number plates. If the algorithm needs to be run with high precision, on the contouring of the number plate in order not to blur the pavement, this is the work that we've been doing on one of our first projects. By working together in a team of data scientists and machine learning engineers, we managed to reduce time-consuming, costly manual work by 92% with AI mechanisms.
We mainly develop on-prem solutions that our clients want to run on their own GPUs, on their own servers; but we also incorporate ready services if needed, using well-known platforms, especially on the Data Engineering, Data Lakes or the Cognitive Services side. In order to support our clients with riding the AI wave, we offer services starting from Data Engineering, to Data Science and Analytics, to Machine Learning.
These services are provided through a structured approach, following a:
- Consultancy phase: discovery with a feasibility study
- Development phase: through prototyping and getting the solution ready for production.
- Monitoring phase: the solution is then constantly monitored through observing the input & output data of a fully functional model.
Error detection and prevention for the Automotive industry
We use AI to bring automation to software, that relies on a certain degree of learning. If this automation can be implemented through rules, then AI would overcomplicate things. For example, if a parameter has a certain value, then a specific action would be triggered. This can easily be implemented through rules mechanisms and AI is not necessary. But when it comes to finding a solution for detecting errors in the entire painting process within an automotive plant, with specific regulations and particular environmental characteristics, an effective AI solution is most certainly required.
This is precisely what we have been working on for one of our projects, deploying a tactical algorithm for error detection. Bluntly put, we build software for ovens used to paint car bodies. These huge ovens need to comply with specific regulations while monitoring the environment's characteristics such as humidity, temperature and dust particles in the air.
Imagine an employee entering such an environment without proper equipment and bringing dust on his/her clothes. These particles could easily get on the car body and in the paint, resulting in painting defects. Consequently, that car body would need to be repainted. To prevent such errors, the AI Team is building algorithms that are learning by themselves which would be the combination of values for humidity, temperature and dust particles in the air, leading to errors in the coating process.
But how do they learn? We use Neural Networks and specific implementations of Python with Mongo DB. These trends are then used to depict any kind of possible match that could lead to painting failures, enabling corrective actions. The feature to error correlation implies continuous monitoring of the qualitative data generated during the painting process. Based on the data provided by the client’s monitoring staff, our algorithms analyze the data, detect if a certain combination of car body features or settings in the paint shop process leads to errors and notify the user about a possible error cause.
At the same time, we develop the algorithms to also detect periods with high error concentration – having the data provided by the monitoring staff, the algorithm would analyze and detect which time frame was prone to having a higher error rate. The impact of our algorithm solutions is undeniable: our client can analyze quality and error data for more than 20.000 car bodies, in time intervals ranging from less than 2 minutes to 10 seconds. The algorithms are executed on a weekly basis and, depending on the data generated in a specific week, the users are notified if a certain feature leads to error or which periods might have higher error rates.
You can find out more about some of our other AI solutions by watching this DEMO on Computer Vision.