The customer is a leading US-based global manufacturer of professional factory-grade systems such as Commercial Air Handlers, Split Heat Pumps, Split Air Conditioners. The company currently operates 100K+ HVAC (heating, ventilating and air conditioning) and water heating systems for its enterprise clients across the globe.
Challenge
Failure in any of the HVAC or water heating systems installed in the field, can have severe impact on operations for enterprise clients. Upon failure, replacement parts or skilled technicians to resolve an issue may not be readily available, leading to further downtime and losses. Our customer wanted the ability to predict system anomalies and automate maintenance requests or alerts for immediate corrective actions.
Solution
Orion developed and implemented an AI/ML-based IoT solution that could track system anomalies and suggest corrective measures. The HVAC and water heating systems installed in the field, regularly report their configuration and sensor measurements to the cloud producing several gigabytes of data per day. The data collected indicates the health and performance of systems and its components, and the solution developed leverages this information to drive alerts regarding predictive maintenance.
The Orion team organized the data received from thousands of sensors, e.g., temperature inside and outside a vessel, power input / output at which a system was running, weather conditions and other information. They then applied Mathematical, Statistical and Deep Machine Learning approaches to the organized datasets and clustered systems to predict failures, their severity and impact on operations.
The Orion team applied the following techniques:
Dynamic Analysis: Each device’s behavior was compared with its past performance and other similar devices. This allowed us to cluster the data and identify devices with anomalous behavior.
Correlation Analysis: We correlated internal measurements with external measurements such as temperature, humidity and more. This helped in identifying device anomalies with respect to external weather conditions.
Classification Analysis: We identified and labeled past events such as alarms and warranty claims, against respective data measured by the sensors at the same instance.
Machine Learning: Deep Machine learning algorithms were trained based on the above analysis and the labeled event data, to predict alarms and warranty claims.
Impact
Orion identified 200+ potentially faulty systems in two months. Along with accurate observations regarding failures, our team delivered a detailed analysis and solution to the customer. We highlighted anomalies such as, an HVAC system starting at full power unnecessarily; and a water heater unable to boil water to the required temperature during certain periods. Our algorithms were iteratively tested in real-world conditions and adapted based on test results.
Finally, our solution was validated by the field engineers, who rectified system issues based on our recommendations. At present, our solution is deployed for continuous monitoring. As a result, system failures have reduced drastically, as most of the issues are flagged-off for predictive maintenance.