This post was co-written with Harsh Vardhan, Global Head of the Digital Innovation Hub, from Apollo Tyres Ltd.

Every recipe change on a tyre extrusion line triggers a stabilization period—minutes of off-spec material that must be reworked or scrapped. For high-mix manufacturers running 30–35 setup changes per day, this adds up to 5–6% of total production volume lost to startup rework. In this post, you learn how Apollo Tyres built an AI-driven APC closed-loop control system to achieve First Time Right tyre extrusion process.

Apollo Tyres Ltd (ATL),headquartered in Gurgaon, India, is a leading global tyre manufacturer with production facilities across India and Europe. As part of its digital transformation journey, ATL established a cloud-native manufacturing data platform on AWS and deploys AI solutions across its manufacturing plants. With ~44% of AI/ML initiatives progressing into production, Apollo Tyres is moving beyond experimentation toward scaled AI adoption across manufacturing.

To tackle one of the most persistent challenges in tyre extrusion—startup instability following recipe changes—Apollo Tyres developed an in-house AI-driven Advanced Process Control (APC) closed-loop solution. The system uses a predictive AI model running on edge to determine the optimal extrusion parameters across 40+ critical and 200+ process features. It also uses an AI agent that monitors tread and sidewall material profiles in real time at downstream and booking scales and automatically adjusts parameters through a closed-loop feedback mechanism to keep material specifications within defined limits.

Unlike conventional feed-forward control systems, this solution employs a real-time closed-loop feedback mechanism with safety guardrails and Human-on-the-Loop (HOTL) oversight.

The APC closed-loop controller compares measured tread and sidewall profiles on every measurement cycle. When a deviation is detected, the controller automatically adjusts the extrusion line speed to bring the material profile back within specification.

This reduced startup rework by approximately 26% and cut stabilization time by approximately 40%, achieving First Time Right (FTR) production.

In this post, we explain how Apollo Tyres Limited (ATL) combines predictive modeling, closed-loop control, and a production-grade machine learning operations (MLOps) pipeline using Amazon SageMaker and Amazon Bedrock to analyze extrusion process data and deliver real-time control decisions. The solution predicts the optimal startup line speed at recipe change and uses closed-loop feedback to track tread and sidewall profiles and adjust line speed when deviations are detected. This approach helps achieve First Time Right (FTR) production while reducing setup and process rework.

The challenge: Can extrusion start at the right line speed—the first time—every time?

The tread and sidewall are critical parts of a tyre because they directly determine road performance. The tread provides grip, braking, and traction, while the sidewall supports the tyre’s structure, absorbs shocks, and maintains stability under load. Even small variations in dimensions or weight can affect tyre safety, durability, ride comfort, and overall performance. Extrusion is the manufacturing process responsible for producing these tread and sidewall components.

Every recipe change—whether driven by a new Stock keeping unit (SKU) compound change or tooling replacement—requires the process to re-stabilize before the required weight and width profiles are consistently achieved. During this phase, the line generates off-spec material that must either be re-worked or scrapped.

At the Apollo Tyres Andhra Pradesh (AP) Plant, Passenger Car Radial (PCR) tyres Extrusion production line operate with 40+ SKUs per month with 30–35 setup changes per day. Every setup change required manual adjustment of extrusion line speed by the operator to bring material profiles within tolerance limits.

The manual approach resulted in:

• Two to Three minutes of stabilization time generating startup rework on every setup
• Weight and width deviations
• Operator dependency
• Re-pass of material through Mixer and Extrusion

This translated into startup rework equivalent to approximately 5–6% of total production volume, increasing the Cost of Poor Quality (COPQ) at the AP Plant PCR line. The impact extended beyond material losses to include higher energy consumption, reduced production capacity, and increased safety risks associated with manual rework handling.

The solution: AI-driven closed-loop control

To eliminate the variability associated with manual trial-and-error startup adjustments, the Apollo Tyres Digital Innovation Hub (DIH) developed an AI-driven APC closed-loop control solution that helps extrusion operators achieve First Time Right (FTR) stabilization after every recipe change. Built on Amazon SageMaker and Amazon Bedrock, the solution combines industrial IoT streaming data (extrusion parameters) with historical process intelligence to predict the optimal startup line speed before extrusion begins. An APC closed-loop control agent evaluates the material profile at each measurement interval and applies corrections when deviations are detected.

The solution analyzes more than 200 features spanning extrusion parameters, downstream systems, and booking measurements across all PCR SKUs and compound configurations. Using predictive and prescriptive analytics, it determines the optimal startup line speed and automatically writes the recommended setpoint to the PLC, reducing startup stabilization time by 40%. The embedded AI agent uses feedback from linear and booking scale measurements and automatically adjusts the extrusion line speed to maintain the material profile within specification, minimizing process corrections and achieving First Time Right (FTR) performance with minimal operator intervention.

Figure 1: Logical flow diagram of Advanced Process Control (APC) Closed-Loop Control process

Beyond process automation, the solution gives engineers a data-driven view of startup instability across machines, recipes, and production shifts.

The solution identifies which process conditions most influence profile deviation. It then recommends the optimal startup line speed at the moment of recipe change, enabling repeatable stabilization across high-volume SKU production.

Key technical capabilities of the APC closed-loop solution include:

• AI-driven startup line-speed prediction using a predictive AI model
• Real-time closed-loop correction using downstream and booking scale feedback
• Real-time streaming analytics for extrusion parameter monitoring
• Automated PLC setpoint updates through edge execution (AWS IoT Greengrass) for low-latency control
• Cloud-native MLOps for model governance, retraining, and scalable deployment
• Explainability dashboards showing First Time Right performance across SKUs, machines, and production shifts.

Business benefits

The measurable outcomes include:

• 26% reduction in setup rework
• 40% reduction in setup stabilization time
• Approx. 20 additional production hours recovered per month
• Improved First Time Right performance

Beyond cost savings, Closed-loop control solution:

• Reduced operator dependency on manual speed adjustments during startup
• Reduced process variability in tread weight and sidewall dimensions across shifts
• Improved workplace safety by minimizing manual handling of rework material
• Delivered repeatable process performance (within specification limits) across machines, shifts, and production teams

ATL tailored structured methodology

Rather than treating this as a standalone AI initiative, Apollo Tyres executed the project within its existing Total Quality Management (TQM) framework and Daily Work Management (DWM) structure. The team applied I5 Agile governance to integrate AI development with domain intelligence from a cross-functional engineering team.

This approach positions AI as an analytical countermeasure to a defined manufacturing problem, not a technology demonstration.

The methodology covers the full lifecycle—from problem identification and solution development to validation, deployment, and continuous improvement. This discipline helps the AI solution deliver measurable business outcomes and supports standardization across Apollo Tyres’ manufacturing operations.

I5 Agile Scrum governance (Initiate, Inspect, Iterate, Improve, Integrate) structures execution through sprints, cross-functional collaboration, and stakeholder reviews. This supports rapid AI experimentation, iterative validation, and controlled deployment into production.

The APSA (Analyze–Prioritize–Set Target–Assign) methodology establishes the business objective and governance structure by:

• Quantifying startup rework losses across the extrusion process
• Defining a SMART improvement target to reduce startup rework (by 25% in FY26)
• Assigning cross-functional ownership across manufacturing, quality, process engineering, and digital teams

The solution was then executed using the OPDCA (Observe–Plan–Do–Check–Act) methodology:

• Observe: Gemba (shop floor) studies identified manual startup line-speed adjustment as the primary source of process instability.
• Plan: Fishbone analysis (Ishikawa diagram) and 5-Why analyses identified manual control, operator variability, and the absence of predictive decision support as key root causes.
• Do: Developed and deployed two components: (1) an AI model running on Amazon SageMaker to predict the optimal startup line speed, and (2) a closed-loop APC controller that monitors the material profile and adjusts line speed in real time.
• Check: Validated performance through statistical before-and-after comparisons against quality metrics, including setup rework rate, and productivity metrics, including stabilization time and First Time Right (FTR) rate.
• Act: Standardized operating procedures, established model governance, and initiated horizontal deployment across applicable extrusion lines.

This methodology keeps the solution technically sound, operationally governed, and ready to scale—while creating a repeatable blueprint for applying AI to future manufacturing challenges at Apollo Tyres.

Technical architecture: APC closed-loop control on AWS

The solution uses a cloud-native architecture on AWS that spans data ingestion (Amazon Kinesis Data Streams, Amazon Data Firehose), storage and transformation (Amazon S3, AWS Glue, Amazon Redshift), predictive modeling and MLOps (Amazon SageMaker), and agentic AI (Amazon Bedrock). At the edge, AWS IoT Greengrass and AWS IoT Core provide low-latency inference and secure MQTT-based PLC communication, while Amazon API Gateway enables authenticated REST and OPC UA integration. Amazon EC2 hosts the solution orchestrator. Security and operational governance are handled by AWS Identity and Access Management (IAM), AWS Key Management Service (AWS KMS), AWS Secrets Manager, AWS CloudTrail, AWS Config, and Amazon CloudWatch. The AI model predicts the optimal startup line speed for each recipe change and writes the setpoint value to the PLC, reducing startup stabilization time and setup rework.

During production, the APC agent monitors tread and sidewall weight and width and closed-loop controller adjusts extrusion line speed in real time to maintain the material profile within specification, keeping off-spec materials in check.

The modular architecture also provides a foundation for future capability like: AI-powered SPC (Statistical Process Control) monitoring that detects process drift using Nelson Rules and recommends corrective actions before defects occur.

Solution workflow

1. The solution analyzes industrial IoT and recipe data available in Amazon Redshift, including historical extrusion parameters, compound information, and environmental conditions.
2. The Amazon SageMaker hosted AI model predicts the optimal startup line speed based on the current recipe, compound, and environmental conditions.
3. The solution automatically writes the predicted line speed setpoint to the PLC through the edge controller. The predicted value is displayed on the HMI for operator oversight.
4. Downstream scales and gauges measure tread weight and sidewall width in real time and send the measurements to the APC controller.
5. The APC closed-loop controller adjusts the extrusion line speed based on the measured profile to maintain the material within specification.

The following diagram illustrates the architecture designed on AWS to achieve the workflow:

Figure 2: Technical architecture diagram of AI-driven APC closed-loop control

The solution uses a modular ,cloud native architecture that can be deployed across additional extrusion lines and manufacturing plants without rebuilding the ML pipeline or edge integration layer.

• Industrial Data Platform: Amazon Kinesis ingests real-time IoT data from PLCs, downstream scales, and recipe system to Amazon S3 and Amazon Redshift, storing and transforming this data into ML-ready features while preserving process causality.
• AI Model and MLOps: Predictive models are developed and managed on Amazon SageMaker, with SageMaker Pipelines and Model Registry for automated training, versioning, validation, and governed deployment.
• Edge AI, Agentic AI, and APC Closed-Loop Control: AWS IoT Greengrass, AWS Lambda, and Amazon API Gateway deliver low-latency edge AI inference. Amazon API Gateway provides secure, authenticated PLC integration via REST and OPC UA protocol. An Amazon Bedrock powered APC agent tracks tread and sidewall profiles and optimizes line speed through closed-loop feedback while operating within safety guardrails and the Human-on-the-Loop (HOTL) framework. This achieves First Time Right production

“For years, extrusion startup losses were accepted as an inevitable part of high-mix manufacturing,” says Harsh Vardhan, Global Head, Digital Innovation Hub, Apollo Tyres Ltd. “We challenged that assumption by moving beyond conventional feed-forward control systems to an AI-powered APC closed-loop control solution built on the fusion of cross-functional manufacturing domain expertise and AI innovation. The AI-powered Advanced Process Control (APC) closed-loop architecture combines predictive Edge AI, APC control agent, and explainable AI. We are achieving First Time Right performance while giving operators greater confidence in every startup decision. Our vision is not to replace operators with AI, but to make AI a force multiplier that augments their expertise, accelerates decision-making, and drives ongoing improvement in manufacturing performance.”

Lessons learned

Apollo Tyres identified the following lessons from this implementation:

• Start with the business problem: Successful AI initiatives begin with clearly defined business outcomes, measurable targets, and strong cross-functional ownership—not technology alone.
• Preserve process causality: Manufacturing AI depends on engineering meaningful features that reflect the physical behavior of the process, not ingesting large volumes of data without context.
• Build trust through phased autonomy: Validating AI recommendations with operators before automating control accelerated adoption and ensured safe, reliable deployment. Starting with operator confirmation (HITL) and graduating to autonomous closed-loop control (HOTL) built trust incrementally.
• Design for scale with MLOps: A governed MLOps framework supports automated model monitoring, versioning, retraining, and scalable deployment across manufacturing lines.
• Combine Predictive AI with Agentic AI: Predictive AI determines the optimal startup line speed, while AI agents explain process deviations, recommend corrective actions, and autonomously adjust line speed through APC closed-loop control—transforming AI from prediction into actionable manufacturing intelligence.

Next steps: Toward self-learning closed-loop manufacturing

The current solution combines AI-driven startup line-speed prediction with real-time APC closed-loop control to achieve First Time Right production. Building on this foundation, Apollo Tyres is evolving the platform toward a self-learning system that monitors, adapts and optimizes manufacturing performance in real time.

The roadmap includes:

• Reinforcement Learning (RL): AI agents learn optimal control policies from live production feedback, allowing the APC system to adapt to changing materials, machine conditions, and operating environments without manual retuning
• AI-powered process monitoring: An SPC agent applies Nelson Rules and monitors Process Capability (Cp & Cpk) to detect process instability early, provide explainable insights, and sustain process capability before quality deviations occur.
• Agentic AI for manufacturing intelligence: Domain-aware AI agents accelerate root cause analysis, recommend corrective actions, and optimize process parameters such as line speed, acting as copilots that augment manufacturing teams with real-time decision support.

Beyond extrusion, Apollo Tyres is developing Apollo Tyres – Manufacturing Reasoner, an internal generative AI solution built on Amazon Bedrock. Manufacturing Reasoner provides manufacturing and enterprise teams with context-aware assistance for operational decision-making.

By combining predictive AI, closed-loop control, reinforcement learning and Agentic AI, Apollo Tyres is laying the foundation for intelligent manufacturing systems that autonomously learn, adapt, and improve operational performance at scale.

Conclusion

In this post, we showed how Apollo Tyres transformed extrusion startup from a manual, trial-and-error process into an AI-driven APC closed-loop control system that achieves First Time Right production. By combining predictive AI on the edge, real-time APC closed-loop control, Amazon SageMaker for MLOps, and Amazon Bedrock for agentic AI — all within a disciplined TQM framework — the solution reduced startup rework by approximately 26%, cut stabilization time by 40%, and recovered approximately 20 additional production hours per month.

This initiative reinforces a broader lesson: successful manufacturing AI requires more than algorithms — it demands structured problem-solving, strong governance, and close collaboration between manufacturing and digital teams. As Apollo Tyres expands the platform with reinforcement learning, AI-powered process monitoring, and multi-line deployment, the vision is clear: AI as a force multiplier for engineers and operators — not a replacement.

To build similar AI-driven manufacturing solutions, explore Amazon SageMaker for predictive modeling and MLOps, Amazon Bedrock for agentic AI capabilities, and AWS for Manufacturing for industry-specific solutions and resources. Share your thoughts and questions in the comments below.