The Challenge
A Fortune 500 specialty chemicals company operates a fleet of 12 batch reactors producing high-value polymer additives. These additives are used in automotive, electronics, and construction applications where consistent quality is critical.
The production process involves a multi-step synthesis with exothermic reactions, precise temperature profiles, and tight quality specifications. Despite having experienced operators and well-tuned PID controllers, the plant faced several challenges:
- Yield variability: Batch-to-batch yields varied by ±3%, impacting profitability
- Off-spec production: 4% of batches required rework or downgrading
- Energy inefficiency: Conservative temperature profiles led to excessive cooling
- Operator dependency: Best yields achieved only by most experienced operators
"We knew there was room for improvement, but our PID-based approach had hit a ceiling. The reaction kinetics are too complex for traditional control to optimize."
— Plant ManagerThe Solution
The plant deployed Acaysia on their highest-volume reactor line, starting with a pilot on two reactors before expanding to all 12 units.
Implementation Approach
- Data collection (4 weeks): Acaysia operated in shadow mode, collecting process data and building hybrid gray-box models that combined first-principles reaction kinetics with neural network components.
- Advisory mode (6 weeks): Operators received real-time recommendations for temperature setpoint adjustments. This phase validated model accuracy and built operator confidence.
- Closed-loop control (ongoing): Acaysia took direct control of temperature profiles, continuously optimizing for yield while respecting all safety constraints.
Technical Highlights
- MPPI-based predictive control with 30-minute horizon
- Real-time model adaptation for feedstock variations
- Integration with existing PLCs via OPC UA
- Seamless failover to PID control with <1 second response time
The Results
After 6 months of operation, the plant documented significant improvements across multiple KPIs:
Yield Improvement
Average yield increased from 91.2% to 93.0%—a 1.8 percentage point improvement. For a product valued at $15/kg with annual production of 80,000 metric tons, this translated to approximately $2.4M in additional revenue.
Quality Consistency
Off-spec batches dropped from 4.0% to 0.8%. The predictive control maintains tighter temperature profiles, resulting in more consistent molecular weight distributions in the final product.
Batch Time Reduction
By optimizing temperature trajectories, average batch time decreased by 12%. This increased effective capacity without capital investment, enabling the plant to meet growing customer demand.
Energy Savings
More precise temperature control reduced cooling water consumption by 18% and steam usage by 9%, contributing an additional $180,000 in annual utility savings.
Before vs. After Comparison
| Metric | Before Acaysia | After Acaysia | Improvement |
|---|---|---|---|
| Average Yield | 91.2% | 93.0% | +1.8 pts |
| Yield Std Dev | ±1.5% | ±0.6% | -60% |
| Off-Spec Rate | 4.0% | 0.8% | -80% |
| Avg Batch Time | 8.2 hours | 7.2 hours | -12% |
| Cooling Water | baseline | -18% | -18% |
Operator Perspective
Initial operator skepticism quickly gave way to enthusiasm as the benefits became apparent:
"At first, I thought it would try to replace us. Instead, it's like having a really smart assistant that watches everything and catches things I might miss. I can focus on the bigger picture while it handles the fine-tuning."
— Senior Reactor Operator, 18 years experienceKey factors in operator acceptance:
- Clear explanation of why recommendations were made
- Easy one-click override at any time
- Gradual rollout from shadow to advisory to closed-loop
- Performance data shared transparently with all shifts
Looking Ahead
Based on the success of this deployment, the company is expanding Acaysia to:
- Two additional production lines at this facility
- Pilot deployment at their European manufacturing site
- Extension to continuous reactor processes (in development)