Optimizing production processes and costs in manufacturing
Optimizing production processes and costs in manufacturing by using Deeploy
Fabric manufacturers all over the world face the same fundamental challenge – how to produce as much output with as few resources as possible. In that regard, Vlisco, a Dutch manufacturer of luxury (clothing) fabrics focusing on the high-income market in Africa, is no exception. Challenged by shrinking margins across the industry while striving to preserve the highest quality of production Vlisco has seen AI as a viable solution.
Balancing efficiency and quality
While a wide spectrum of solutions could be applied to this challenge, it starts with one simple question. How to produce the required products in the most efficient way while still reaching & maintaining quality standards?
The answer to that is far more complex than to be captured in a few basic business rules and is a challenge that Vlisco is trying to solve by using Machine Learning.
‘Predictive material input’ and ‘Production error detection’ and the challenges for Vlisco
While anomaly detection has been a manual process in the past, Machine Learning opens up new opportunities. With the support of Machine Learning, Vlisco is able to detect anomalies in their fabrics by processing large numbers of production output quickly and consistently.
With a complex production process, intensive labor activity, and being located in the Netherlands, Vlisco deals with comparatively high labor costs. To optimize the effectiveness of its operation, Vlisco aims to leverage Machine Learning to detect fabric anomalies and maintain quality standards while keeping costs low.
Vlisco built Machine Learning models that utilize – “Predictive material input” and “Production error detection”-both pursuing the goal of optimizing material usage.
On the first look, it looks like a fairly simple model. And, yet, it heavily relies on the availability of production data as well as product & production order characteristics in order to correctly function.
Moreover, to deliver the expected high quality of Vlisco’s products to customers, the Machine Learning models must continuously prove their high accuracy.
Image: Example of AI-supported anomaly detection at Vlisco
As such, while the theoretical benefit might be obvious, Vlisco faced several challenges in the Machine Learning adoption process.
First, they had to design and set up their entire Machine Learning infrastructure. Second, their ambition to scale their ML application across their complete product catalog, presented a further challenge . Third, company-wide acceptance was needed to fully use the potential of ML usage in their daily operations. Lastly, and due to the continuous development of innovative new designs in their production process, AI models need to be trained by receiving feedback from Vlisco’s personnel.
Use case ‘Reduction of material excess’
Summary
- Ambition: Optimize required raw inputs for desired outputs of products to improve production costs
- Situation: Due to the complexity and manual nature of the production process, % of input materials ends up as scrap which increases cost significant
- ML model: input material estimation based on historical production data and order characteristics
Challenges
- Potential future degradation of performance due to changing factory environment and designs
- Highly manual process and limited experience with ML models or applications
- ML infrastructure still in the early stages of set up
Deeploy’s platform: manageable AI supported by human feedback at Vlisco
As the desired AI models directly impact existing business processes, performance, explainability, and trust were key for successful adoption. Therefore, a close collaboration between humans and production machines was essential, a perfect fit with Deeploy’s platform.
In addition to providing a user-friendly, scalable platform for model deployment and management, Deeploy offers a unique feature: the human feedback loop. This feedback loop allows Vlisco’s business experts to directly evaluate and, if necessary, overrule predictions.
This, not only creates trust and transparency for experts and users but also allows for giving real-time feedback that can be used to retrain the models. Thus, each prediction improves the estimation of how much material is required to produce a certain output, directly influencing Vlisco’s production costs.
Moreover, to ensure model performance, Deeploy provides the option to set customized, real-time alerts, informing the Data Science Team about performance drifts and allowing them to take necessary corrective measures. This is especially important when it comes to ensuring the highest accuracy possible for anomaly detection models. Based on the alerts, Vlisco’s team is able to improve and deploy new, improved models without risking any downtime.
Another interesting feature are audit trails. Every model prediction is logged on Deeploy and Vlisco is easily able to reproduce and document any prediction the model made, ensuring full model accountability.
Deeploysupports Vlisco as a solution that creates full explainability for every model, while simultaneously allowing for receiving feedback that can be used for model improvement. A platform ready to be scaled for new use cases and models
Want to explore the full potential of Deeploy?
