Cement and Mining Equipment Manufacturer Reduces Unplanned Downtime with AI Insights

Challenges

A global manufacturer is a leading supplier of cement and mining equipment such as pumps, crushers, mills, high-pressure grinding rolls (HPGR), and other industrial equipment, with customers in over 150 countries. A top priority for the company is ensuring its customers maximize value from installed assets by improving their operational efficiency and avoiding equipment downtime.

The company’s monitoring teams and subject matter experts provide advisory and monitoring services, delivering operational advice and machine condition reports to customers. However, the company’s existing advising and condition report solutions had several limitations:

• Historical sensor and KPI dashboards had little predictive capability
• Data science and engineering teams required 4 days to onboard and deploy ML models for new assets
• Data science teams had limited ability to add value through additional machine learning experiments, given the time required for asset onboarding and the solution’s lack of predictive power
• The monitoring support team lacked tools to deploy and manage live production models at scale or to provide access to end customers who want to use the tools to monitor their machines

Approach

Over 24 weeks, C3 AI partnered with the manufacturer to configure C3 AI Reliability and enable predictive maintenance across customer operations. The team began by ingesting, cleansing, and unifying two years of historical time series data across four asset classes and 2,600+ sensors. The company uses a Snowflake data lake for sensor data, so the team configured C3 AI Reliability to integrate directly with Snowflake for model training and inferences, as well as plotting and querying data.

After integrating data from 62 machines, the joint team performed exploratory data analysis and identified high variability in sensor data. Since the manufacturer does not directly operate the machines it produces, it did not have access to downtime and maintenance data. To address these challenges, the C3 AI team designed a modeling approach based entirely on sensor data. The team applied unsupervised anomaly detection algorithms and validated results with SMEs during development using the application’s model feedback user workflows.

The application enabled the manufacturer to provide early warning of critical failure modes, such as seal failure, improper lubrication, cavitation, and high vibration. Furthermore, C3 AI Reliability provided prescriptive corrective actions and detailed data-driven evidence packages to accelerate engineering troubleshooting and risk management. ML models on pumps and other asset classes demonstrated precision of 87%, defined as the proportion of useful alerts to total alerts.

About the Company

• $3+ billion annual revenue in 2023
• 60+ offices across 150 countries
• 11,000+ employees

Project Objectives

• Enable predictive maintenance to accurately predict failures across 4 asset types
• Integrate and unify data from 62 machines and 2,600 sensors to unlock machine learning insights for customers
• Enhance monitoring and advisory services to increase asset performance and availability
• Configure the C3 AI Reliability application to surface AI insights in an intuitive interface

Project Highlights

• 24 weeks to configure an enterprise-grade, production application
• 200GB live data sources integrated, including Snowflake data virtualization and data from 62 machines and 2,600 sensors
• 4 asset types monitored with a flexible, extensible data model
• 25+ customer data scientists and operators trained to onboard assets and use the application
• 12-month scale out plan for company to deploy C3 AI Reliability across global customer operations
• 6 C3 AI Reliability screens configured, with user interface customized to the manufacturer’s business