Novel AI Predictive Maintenance Delivers $5.3M ROI in Just 30 Days

This success story highlights EagleAPM, a predictive maintenance platform conceived and architected by Chirag Agrawal, Global Head of Data Science at Novelis, and built in partnership with Konverge AI. EagleAPM is a core part of Novel AI, Novelis’ AI-driven predictive maintenance initiative designed to improve asset reliability and enable smarter plant operations. Chirag Agrawal led the end-to-end architecture for EagleAPM, shaping the platform around three capabilities that set it apart from conventional predictive maintenance solutions.

3 capabilities which make EagleAPM different

  • Unified anomaly detection: Combines statistical models, machine learning, computer vision, and acoustic analysis in one platform
  • Built for OT environments: Designed to operate within industrial security and connectivity constraints
  • Self-service for reliability teams: Enables plant engineers to configure and manage predictive alerts without depending heavily on data science teams

We worked closely with Chirag Agrawal and his team to translate this vision into a production-ready solution aligned to operational and plant-level requirements. This also helped shape Konverge AI’s broader approach to industrial predictive maintenance solutions.

Novel AI Predictive Maintenance Delivers $5.3M ROI in Just 30 Days

This success story highlights EagleAPM, a predictive maintenance platform conceived and architected by Chirag Agrawal, Global Head of Data Science at the world’s largest aluminum rolling and recycling company, and built in partnership with Konverge AI. EagleAPM is a core part of Novel AI, an AI-driven predictive maintenance initiative designed to improve asset reliability and enable smarter plant operations. Chirag Agrawal led the end-to-end architecture for EagleAPM, shaping the platform around three capabilities that set it apart from conventional predictive maintenance solutions.

3 capabilities which make EagleAPM different

1

Unified anomaly detection

Combines statistical models, machine learning, computer vision, and acoustic analysis in one platform

2

Built for OT
environments

Designed to operate within industrial security and connectivity constraints

3

Self-service for reliability teams

Enables plant engineers to configure and manage predictive alerts without depending heavily on data science teams

We worked closely with Chirag Agrawal and his team to translate this vision into a production-ready solution aligned to operational and plant-level requirements. This also helped shape Konverge AI’s broader approach to industrial predictive maintenance solutions.

The Industry Challenge

Despite heavy investment in industrial IoT, production-scale predictive maintenance remains rare in heavy manufacturing. The main reason being architectural barriers (not algorithmic).
  • OT (Operational Technology) Security Constraints: Commercial APM platforms (PTC ThingWorx, GE Predix, Siemens MindSphere), were built for the cloud. Heavy manufacturing environments operate under strict Purdue Model network segmentation that prohibits direct sensor-to-cloud data flows.
  • Alert Fatigue: Conventional systems rely on single detection methods, resulting in excessive false alarms across equipment types and failure modes. Operators learn to distrust/ignore the system.
  • Data Science Dependency: Most platforms require trained data scientists for ongoing model configuration, creating a permanent bottleneck that prevents scaling beyond the pilot stage.

Our Solution

EagleAPM addresses all three challenges through a fundamentally different design:
  • Hybrid Multi-Modal Alerting: Five distinct alert types—static threshold, dynamic statistical, ML-based (supervised and unsupervised), computer vision, and acoustic anomaly detection, operate within a unified framework. Each is deployed where it works best, reducing false positives without missing any real failures. A built-in feedback loop keeps the system self-correcting. No commercial platform offered this combination of five detection modalities with closed-loop feedback.
  • OT-Native Deployment: Built for Purdue Model industrial environments, EagleAPM processes sensor data locally within plant OT infrastructure and connects with enterprise systems through approved integration pathways. Its plant-agnostic architecture supports varying OT stacks, sensor configurations, historian systems, and network topologies, enabling scalable deployment across facilities in the United States and Canada without requiring fundamental redesign.
  • Self-Service Configuration: Reliability engineers configure alerts, set asset thresholds, assign routing rules, and manage performance through a simple interface, no coding, no data science support. Once deployed, local plant teams operate and extend the platform independently, eliminating the scalability bottleneck inherent in conventional APM solutions.

Business Benefits

We implemented EagleAPM at multiple sites and machine centers. The primary site achieved $5.3 million in verified operational benefits within the first 30 days of production deployment, driven by early detection of three critical equipment anomalies that would have resulted in unplanned downtime costing an estimated $1.5–$2.0 million per incident.

Following the success of EagleAPM, Konverge AI adopted its core architectural patterns as a reference design for industrial predictive maintenance solutions.

  • EagleAPM’s Three Architectural Principles: We adopted EagleAPM’s three architectural patterns as the standard reference architecture for industrial predictive maintenance solutions. Mr. Chirag Agrawal’s original architecture now underpins Konverge AI’s broader industrial AI solution approach and is being adapted beyond the aluminum sector.
  • Redefining Industrial AI: This engagement evolved Konverge AI’s industrial AI offerings towards a more scalable, plant-ready operating model. EagleAPM’s hybrid alerting framework marked a clear advancement over traditional single-method APM implementations.
  • Three Core Challenges EagleAPM Addresses: Heavy manufacturing — steel, chemicals, cement, paper, and glass all face identical OT security, alert fatigue, and data science dependency barriers. Konverge AI’s adoption of EagleAPM as a reference design is initial evidence of this broader applicability.

EagleAPM sets a new benchmark for industrial predictive maintenance by combining a differentiated architectural approach with proven performance and relevance beyond its original deployment.

Download Detailed Case Study

DOWNLOAD NOW

Related Resources

Case Studies

{{item.authore_images_alt}}

Vehicle damage detection using Computer Vision for insurance company

Our customer, a born-digital unicorn in Insurance industry wanted to use Computer Vision for automation.
{{item.authore_images_alt}}

Recommendation engine development for a Media Firm

Driving multifold improvement in recommendation efficiency Our customer in Media Distribution...
{{item.authore_images_alt}}

Digitizing Electronic Health Records and Making them searchable through Natural Language Processing

NLP based digitization of Electronic Health Records (EHR)
Share this...

To provide your business a competitive edge with our AI solutions

Contact Us