📋 Case Study

Analytical Process Monitoring in Challenging Environments

Maintaining calibration stability and measurement accuracy of inline Raman spectrometers under extreme ambient conditions: ambient temperatures up to 55°C, high solar radiation, corrosive H₂S/SO₂-laden atmosphere, and mechanical vibration from adjacent compressors — all causing spectral drift, detector noise inflation, and fiber-optic coupling degradation.

🏗️ Project Overview

Real-time analytical process monitoring system deployed at a petrochemical refinery in Jubail Industrial City, Saudi Arabia; serving a continuous hydrodesulfurization (HDS) unit processing 120,000 barrels per day of sour crude feedstock.

🎯 Challenge

🔧 Design Approach

Hybrid physics-informed chemometric design: (1) Active thermal stabilization of spectrometer core using Peltier-controlled enclosure (±0.1°C), (2) Vibration-dampened optical mounting with kinematic alignment, (3) Dual-reference calibration strategy combining NIST-traceable gas cells and in-situ flow-through quartz cuvettes with automated baseline correction via constrained alternating least squares (cALS), (4) Edge-computing preprocessing pipeline implementing real-time Savitzky-Golay denoising and multiplicative scatter correction (MSC) before PLS regression.

📐 Design Diagram

• H₂S/SO₂ corrosion
• Solar radiation
• Vibration (5.2 mm/s) Raman Spectrometer Core (785 nm) Peltier ±0.1°C Kinematic Mount NIST Gas Cell Flow Cuvette Edge Preprocessing SG + MSC → PLS Key Performance Bounds Δλ = 0.024 nm SNR: 185 → 94 ε_transfer ≤ 0.83 wt% S

AI-generated project design illustration

📐 Key Calculations

Thermal Drift Compensation Margin

Δλ = α × ΔT × λ₀

Result: 0.024 nm (for λ₀ = 785 nm, α = 0.012 nm/°C, ΔT = 20°C)

Quantifies maximum expected Raman peak shift due to uncontrolled temperature rise; informed the design of the ±0.1°C thermal control spec to limit shift to <0.001 nm—well below spectral resolution (0.05 nm).

Signal-to-Noise Ratio (SNR) Degradation Factor

SNR_degraded = SNR_baseline / √(1 + (k × V²))

Result: SNR_degraded = 185 → 94 (at 5.2 mm/s RMS vibration)

Predicted SNR loss from measured compressor-induced vibration justified implementation of passive damping and lock-in amplification, restoring effective SNR >170.

Calibration Transfer Error Bound

ε_transfer ≤ ||X_target − X_reference||_F × ||B||_F

Result: ε_transfer ≤ 0.83 wt% sulfur (95% CI)

Provided statistical confidence for deploying a single master calibration model across six identical HDS trains, eliminating need for per-unit recalibration and reducing downtime by 70%.

📊 Results

Metrics: Sulfur concentration RMSE: 0.11 wt%, System uptime: 99.3% over 18 months, Calibration interval extended from 48 to 336 hours, Operator intervention events reduced by 82%

The system achieved sub-minute compositional feedback for HDS effluent sulfur content with laboratory-equivalent accuracy, enabling closed-loop catalyst activity compensation and reducing off-spec product generation by 94%.

💡 Lessons Learned

•Environmental hardening must be co-designed with chemometric modeling—not retrofitted
•Vibration-induced mode coupling in fiber optics requires modal analysis early in sensor integration
•Edge-based spectral preprocessing reduces latency and network load more effectively than cloud-only approaches

✅ Key Takeaways

1Robust analytical process monitoring in harsh environments demands concurrent optimization of hardware resilience, optical stability, and adaptive multivariate calibration—not just sensor selection.