Leveraging AI and ML for Predictive Maintenance and Paper Manufacturing Company

The client is a big paper producer, but faced challenges because of expensive maintenance and machines breaking down unexpectedly.

Minor maintenance was carried out every six years at a cost of $3.4M, and major maintenance was performed every 12 years, costing around $7M. Additionally, the company had to halt operations for ten days every year to meet regulatory requirements. Therefore, they reached out for AI consulting services.

Urgent and expensive maintenance: The company faced challenges in predicting which machines required maintenance in advance, resulting in the need for urgent and more expensive maintenance services.

Third-party managed solution: The company attempted to resolve the issue by implementing a third-party managed solution for some machines. However, the high costs associated with this solution prevented scalability across the entire factory.

Engagement of an IT consulting and services provider: To address the challenge, the company engaged an experienced IT consulting and services provider.

The consulting firm used AI and ML to help the company analyze machine behavior and identify small problems early on.

Accurate and efficient maintenance: By detecting minor issues in advance, the company was able to perform maintenance more accurately and efficiently during the obligatory ten-day shutdowns.

Increased interval for major maintenance: The implementation of AI and ML technologies allowed for an increased interval between major maintenance activities, improving overall productivity and performance for the manufacturer.

• Comprehensive data exploration: The partner conducted a thorough exploration of the data's characteristics before commencing the project.
• Highly scalable solution: AWS SageMaker was utilized to devise a scalable solution that enabled the application of tailored predictive models to multiple machines within the factory. This innovative approach enhanced cost efficiency and eliminated costly and non-scalable third-party contracts.
• Thorough data preprocessing: Data was preprocessed comprehensively to prepare it for the DeepAR model.
• Effective model selection: The most effective model was identified through model selection based on the mean absolute error metric.
• DeepAR forecasting algorithm: A deep learning model utilizing the DeepAR forecasting algorithm was implemented to generate forecasts for multiple time series using 92 different time series sourced from various factory sensors.
• Our partner undertook a meticulous fine-tuning process to systematically explore different hyperparameters and enhance model performance.
• After training, a flexible pipeline was created to make it easier to create models for various sensors and machines.
• Model retraining: The pipeline facilitates model retraining, ensuring that forecasts become increasingly precise over time.
• Continuous improvement: e-Core's implemented pipeline supports the iterative process of continuous improvement for machine learning models.

• Increased operational efficiency: Predictive maintenance minimizes downtime by allowing for the prediction of sensor values and proactive maintenance.
• This leads to a longer lifespan for machinery, reducing downtime and boosting productivity, resulting in significant long-term cost savings.
• Implementing AI strategy like predictive maintenance extends maintenance intervals, resulting in an extended equipment lifespan. Maintenance that used to occur every 6 years can now occur every 8 years. Additionally, maintenance that used to occur every 12 years can now occur every 15 years. This can yield cost savings of up to almost $260,000 for a single piece of equipment.
• Less unplanned downtime: Accuracy rates of 80% or more in predicting maintenance needs greatly reduce unplanned downtime. Conservative estimations suggest a reduction of about 20% to 40%.
• Financial impact and ROI: Companies embracing predictive maintenance can expect significant cost savings. For a project with an investment cost of around $70,000, the company forecasts cost-savings of almost $850,000 per year. The investment payback is estimated to occur within 6-8 months.