Transform maintenance with predictive analytics

Production stoppages are a significant headache in the chemical industry. Just one unplanned halt can destabilize the entire operation. This was the problem faced by a leading German chemical company. The culprit was scaling—an unwanted material build upon reactor surfaces that stopped production and increased costs. The challenge was clear yet critical: predict when scaling would reach a level requiring maintenance and do so early enough to avoid an unplanned shutdown.

‍
The Challenge: Avoiding Unplanned Shutdowns

The reactor, a vital part of the production process, would develop scaling over time, reducing efficiency until a complete shutdown was needed for cleaning and maintenance. Although the frequency of these stoppages is low, their impact is substantial, and knowing when one might occur is necessary for the risk of increased costs to stay manageable.

There was a need for a way to anticipate this problem and plan reactor maintenance, ideally months in advance. This would allow the company to schedule downtime on its terms, minimizing the number of unplanned stops and the chaos they cause.

The Solution: Leveraging Real-Time Data and Intelligent Algorithms

AG Solution's team of experts developed a solution aimed at making a seemingly unpredictable problem predictable:

1. Real-Time Data Integration: Our solution connects with the company's AVEVA PI system, utilizing real-time data from the reactor. This system monitors 181 variables, including temperature, pressure, and water injection rates.
2. Anomaly Detection: Given the vast volume of data, a method was needed to filter and identify key indicators of potential scaling. Our experts developed an anomaly detection algorithm using Isolation Forest, focusing on the 30 most critical variables out of the initial 181.
3. Proactive Maintenance Alerts: Once the AI model detected an anomaly, it sent notifications through the AVEVA PI system and via email, informing the team that a scaling issue might be approaching. This allowed them to plan maintenance up to six months in advance.

‍

The Results: Predictive Maintenance that Keeps Production Running

The impact of this solution was almost immediately apparent:

• Continuous Monitoring: The system continuously monitors the reactor's key variables to detect any early signs of trouble. This constant vigilance enables proactive scaling management before it becomes critical.
• Improved Operational Efficiency: The company significantly improved its operational efficiency by shifting from a reactive to a proactive approach. Maintenance can be scheduled during planned shutdowns, avoiding interruptions and consistently meeting production targets.
• ‍Understanding Scaling Causes: The AI model provides insights into the parameters that most affect scaling accumulation in the reactor, highlighting which variables most impact the generation of anomalies.

‍

Conclusion: Setting a New Standard for Predictive Maintenance

This project illustrates how predictive maintenance can revolutionize industrial operations. By combining real-time data with intelligent algorithms, companies can anticipate problems before they occur, keeping production lines running and avoiding costly downtime.