# Continuous PAT & Real-Time Release Testing (RTRT) in Flow Skids: Sensor Fusion, Chemometrics & ICH Q13 Compliance

# Executive Summary & Regulatory Context

In traditional batch pharmaceutical manufacturing, quality assurance relies heavily on off-line quality control (QC) testing of isolated intermediates and end-product API samples. Samples are withdrawn, sent to an analytical testing laboratory, and analyzed via offline HPLC/GC. This offline paradigm introduces feedback delays ranging from hours to weeks, holds large inventory volumes in quarantine, and provides zero real-time visibility into process dynamics.

Continuous Process Analytical Technology (PAT) and Real-Time Release Testing (RTRT) transform quality assurance from reactive testing to proactive, continuous control. By embedding high-speed inline and online analytical sensors—such as Diamond Attenuated Total Reflection FTIR (ATR-FTIR), benchtop 60 MHz quantitative NMR ( $q\text{NMR}$ ), and ultra-fast HPLC—directly into continuous flow skids, process engineers measure Critical Quality Attributes (CQAs) continuously at steady state.

Under ICH Q13 guidelines (Continuous Manufacturing of Drug Substances and Drug Products), a validated RTRT system combined with automated diverter valve logic allows real-time material release, bypassing traditional off-line release testing.

This guide details the optical physics, chemometric algorithms, sensor fusion architectures, and regulatory frameworks required to implement RTRT on industrial continuous flow skids.

# 1. Architectural Overview of Continuous PAT & Sensor Fusion

A continuous PAT sensor network integrates multiple orthogonal analytical detectors along the continuous flow path:

                      CONTINUOUS PAT & SENSOR FUSION SCHEMATIC
 ┌─────────────────────────────────────────────────────────────────────────────────────────────┐
 │                                                                                             │
 │  ┌──────────────┐     ┌──────────────┐     ┌──────────────┐     ┌──────────────┐            │
 │  │ Flow Reactor │────►│ PAT Node 1:  │────►│ PAT Node 2:  │────►│ PAT Node 3:  │────┐       │
 │  │ (SiC / PFR)  │     │ Diamond      │     │ Benchtop     │     │ Rapid UPLC   │    │       │
 │  │              │     │ ATR-FTIR     │     │ 60 MHz NMR   │     │ (90s Cycle)  │    │       │
 │  └──────────────┘     └──────────────┘     └──────────────┘     └──────────────┘    │       │
 │                              │                    │                    │            │       │
 │                              ▼                    ▼                    ▼            │       │
 │                       ┌────────────────────────────────────────────────┐            │       │
 │                       │ CHEMOMETRIC ENGINE & SENSOR FUSION (PLS / PCA) │            │       │
 │                       │ Calculates: Yield (%), Conversion, Impurities  │            │       │
 │                       └───────────────────────┬────────────────────────┘            │       │
 │                                               │                                     │       │
 │                                               ▼                                     ▼       │
 │                       ┌────────────────────────────────────────────────┐     ┌──────────────┐
 │                       │ DCS CONTROL & RTRT DECISION ENGINE             │────►│ 3-Way OOS    │
 │                       │ Validates Specs Against ICH Q13 Criteria       │     │ Diverter     │
 │                       └────────────────────────────────────────────────┘     └──────────────┘
 └─────────────────────────────────────────────────────────────────────────────────────────────┘

# 1.1 Comparison of Continuous PAT Technologies

                  PAT ANALYTICAL DETECTOR SELECTION MATRIX
 ┌──────────────────────┬────────────────────────┬────────────────────────┬────────────────────────┐
 │ PAT Detector         │ Sampling Frequency     │ Limit of Detection (LOD)│ MONITORED PARAMETERS   │
 ├──────────────────────┼────────────────────────┼────────────────────────┼────────────────────────┤
 │ Diamond ATR-FTIR     │  $1 - 10\,\text{s}$      │  $0.1\,\text{wt\%}$      │ Reaction kinetics, functional group conversion│
 │ Benchtop 60 MHz NMR  │  $15 - 60\,\text{s}$     │  $0.05\,\text{wt\%}$     │ Molar ratio, regioisomers,  $q\text{NMR}$  assay│
 │ Rapid Inline UPLC    │  $90 - 180\,\text{s}$    │  $0.005\,\text{wt\%}$    │ Trace impurities, enantiomeric purity│
 │ Transmission UV-Vis  │  $0.1 - 1.0\,\text{s}$   │  $0.01\,\text{wt\%}$     │ Chromophore concentration, dispersion│
 │ Inline Raman (785 nm)│  $5 - 30\,\text{s}$      │  $0.2\,\text{wt\%}$      │ Polymorphism, crystallization phase  │
 └──────────────────────┴────────────────────────┴────────────────────────┴────────────────────────┘

# 2. Optical Physics & Analytical Cell Mechanics

# 2.1 Diamond ATR-FTIR Evanescent Wave Optics

Attenuated Total Reflection FTIR utilizes a high-refractive-index diamond crystal ( $n_1 = 2.42$ ) in contact with the liquid flow stream ( $n_2 \approx 1.35 - 1.45$ ). When infrared light strikes the interface at an angle $\theta > \theta_{\text{critical}}$ , total internal reflection occurs, generating an Evanescent Wave that penetrates into the liquid stream:

                DIAMOND ATR-FTIR EVANESCENT WAVE MECHANICS
 ┌─────────────────────────────────────────────────────────────────────────┐
 │ Liquid Flow Stream (n2)                                                 │
 │ ──────────────────────~ Evanescent Wave ~───────────────────────────── │
 │ ═══════════════════════════════════════════════════════════════════════ │
 │ Diamond Crystal (n1 = 2.42)                                             │
 │      \                                                         /        │
 │       \ IR Light In (θ > θ_crit)             IR Light Out /         │
 └─────────────────────────────────────────────────────────────────────────┘

The Penetration Depth ( $d_p$ ) of the evanescent wave is given by:

d_p = \frac{\lambda}{2 \pi \cdot n_1 \cdot \sqrt{\sin^2\theta - \left( \frac{n_2}{n_1} \right)^2}}

where:

$\lambda$ is IR wavelength ( $\mu\text{m}$ ),
$\theta$ is angle of incidence ( $45^\circ$ ),
$n_1, n_2$ are refractive indices of diamond and liquid medium.

For typical mid-IR wavelengths ( $\lambda = 5 - 10\,\mu\text{m}$ ), $d_p \approx 0.5 - 2.0\,\mu\text{m}$ . Because penetration depth is extremely shallow, ATR-FTIR flow cells are immune to path-length clogging and turbidity issues, making them ideal for highly concentrated API reaction streams.

# 2.2 Benchtop 60 MHz Quantitative NMR ( $q\text{NMR}$ )

Continuous benchtop NMR flow cells ( $5\,\text{mm}$ glass flow tube) utilize permanent NdFeB magnet arrays operating at $60\,\text{MHz}$ ( $1.4\,\text{Tesla}$ ).

The mole fraction ( $x_A$ ) of target API intermediate $A$ relative to starting material $B$ is determined directly from integrated proton resonance intensities ( $I_A, I_B$ ) without requiring empirical calibration curves:

x_A = \frac{\frac{I_A}{N_{\text{H}, A}}}{\frac{I_A}{N_{\text{H}, A}} + \frac{I_B}{N_{\text{H}, B}}}

where $N_{\text{H}, A}$ and $N_{\text{H}, B}$ are the number of protons responsible for the respective NMR resonance peaks.

# 3. Chemometrics & Multivariate Calibration Modeling

Raw PAT spectral data (FTIR absorbance matrices or Raman spectra) contain overlapping peaks, baseline tilt, and temperature-induced scattering. Converting raw spectra into concentration estimates requires multivariate chemometric algorithms.

                  CHEMOMETRIC DATA PROCESSING PIPELINE
 ┌─────────────────────────────────────────────────────────────────────────┐
 │ 1. RAW SPECTRA  ──► 2. PREPROCESSING ──► 3. PLS MODEL ──► 4. OUTPUT     │
 │ (1000s Wavenumbers)  (SNV + 2nd Deriv)    (Y = X * B)    (Yield % & CQA)│
 └─────────────────────────────────────────────────────────────────────────┘

# 3.1 Spectral Preprocessing Protocols

Standard Normal Variate (SNV): Eliminates baseline shifts caused by light scattering or bubble passage:

x_{i, \text{SNV}} = \frac{x_i - \bar{x}}{\sigma_x}

Savitzky-Golay 2nd Derivative: Resolves overlapping absorption bands and removes constant linear baseline offsets.

# 3.2 Partial Least Squares (PLS) Regression Modeling

Partial Least Squares (PLS) decomposes both the spectral predictor matrix $\mathbf{X}$ ( $m \text{ samples} \times p \text{ wavenumbers}$ ) and the response matrix $\mathbf{Y}$ ( $m \text{ samples} \times q \text{ CQAs}$ ) into latent variable factors:

\mathbf{X} = \mathbf{T} \mathbf{P}^T + \mathbf{E}

\mathbf{Y} = \mathbf{U} \mathbf{Q}^T + \mathbf{F}

where $\mathbf{T}$ and $\mathbf{U}$ are score matrices, $\mathbf{P}$ and $\mathbf{Q}$ are loading matrices, and $\mathbf{E}, \mathbf{F}$ are residual error matrices.

The optimal number of latent variables ( $LV$ ) is selected using cross-validation to minimize the Root Mean Square Error of Cross-Validation ( $\text{RMSECV}$ ):

\text{RMSECV} = \sqrt{\frac{\sum_{i=1}^N (y_i - \hat{y}_{i, \text{CV}})^2}{N}}

                 RMSECV VS. LATENT VARIABLES (LV) SELECTION
  RMSECV
    ▲
    │   \
    │    \  Overfitting Zone (High LVs)
    │     \         /
    │      \_______/   ◄ Minimum RMSECV (Optimal LVs = 4)
    └────────────────────────────────────────► Number of Latent Variables (LV)

# 3.3 Anomaly Detection: Hotelling's $T^2$ & $Q$ -Residuals

To ensure that the PLS model is not extrapolating outside its validated calibration space, the chemometric engine computes two real-time multivariate health metrics:

Hotelling's $T^2$ : Measures variation within the model space:

T^2 = \mathbf{t}_i \mathbf{S}^{-1} \mathbf{t}_i^T \le T^2_{\text{crit, 95\%}}

$Q$ -Residuals (Squared Residual Payload): Measures variation outside the model space (e.g., unexpected impurity or solvent contamination):

Q_i = \mathbf{e}_i \mathbf{e}_i^T \le Q_{\text{crit, 95\%}}

If either $T^2$ or $Q$ exceeds the $95\%$ confidence limit, the PAT engine flags a Multivariate Anomaly, invalidating the RTRT prediction and triggering an alert.

# 4. ICH Q13 Regulatory Control Strategy & Diverter Valve Automation

Under ICH Q13, a compliant continuous manufacturing control strategy requires three distinct levels of quality control:

                 ICH Q13 THREE-LEVEL CONTROL STRATEGY
 ┌─────────────────────────────────────────────────────────────────────────┐
 │ LEVEL 1: Real-Time Automatic Feedback / Feedforward Control            │
 │ (PAT sensors adjust pump flow rates & temperatures dynamically)         │
 ├─────────────────────────────────────────────────────────────────────────┤
 │ LEVEL 2: In-Process Controls (IPC) & Out-of-Spec (OOS) Diversion       │
 │ (Automated 3-way diverter valve routes OOS material to waste tank)      │
 ├─────────────────────────────────────────────────────────────────────────┤
 │ LEVEL 3: End-Product Testing & Certificate of Analysis (CoA) Release   │
 │ (RTRT surrogate models replace offline release testing)                │
 └─────────────────────────────────────────────────────────────────────────┘

# 4.1 Out-of-Spec (OOS) Material Diversion Dynamics

When an inline PAT detector measures a CQA violation (e.g., active product purity $< 98.0\%$ ), the DCS system must divert the un-compliant fluid volume before it enters downstream crystallizers or collection drums.

The total volume diverted ( $V_{\text{divert}}$ ) is calculated using the Residence Time Distribution (RTD) of the system:

V_{\text{divert}} = Q_{\text{total}} \cdot \left( t_{\text{PAT\_lag}} + 3 \cdot \tau_{\text{downstream}} + t_{\text{valve\_response}} \right)

               AUTOMATED DIVERTER VALVE TIMING DIAGRAM
 ┌───────────────────────────────────┬───────────────────┬───────────────────┐
 │ Phase                             │ Duration          │ System Status     │
 ├───────────────────────────────────┼───────────────────┼───────────────────┤
 │ PAT Measurement & Chemometrics    │  $2.0\,\text{s}$    │ CQA Breach Detected│
 │ DCS Signal & Valve Actuation      │  $< 50\,\text{ms}$   │ Diverter Valve Switched│
 │ Transient Out-of-Spec Diversion   │  $3 \cdot \tau$     │ Fluid Routed to Waste Tank│
 │ Steady-State Re-establishment     │  $C_{\text{PAT}}$  Stable│ Diverter Valve Reset to Main│
 └───────────────────────────────────┴───────────────────┴───────────────────┘

# 5. Industrial Case Study: RTRT Deployment on a Commercial $50\,\text{kg/day}$ API Skid

# 5.1 System Configuration & PAT Deployment

A commercial flow skid performing a continuous $S_NAr$ reaction followed by quenching was equipped with an integrated PAT suite:

Node 1 (Post-Reactor): Inline Diamond ATR-FTIR ( $10\,\text{s}$ scan interval) tracking reactant consumption at $1520\,\text{cm}^{-1}$ .
Node 2 (Post-Quench): Benchtop 60 MHz $q\text{NMR}$ ( $30\,\text{s}$ scan interval) tracking product purity and residual amine stoichiometry.
Node 3 (Final Liquid Stream): Automated rapid UPLC ( $120\,\text{s}$ cycle time) assaying trace regioisomer impurities.

# 5.2 Results & Commercial Benefits

              COMMERCIAL RTRT VS. TRADITIONAL QC COMPARISON
 ┌───────────────────────────────────┬───────────────────┬───────────────────┐
 │ Metric                            │ Traditional Offline QC│ Integrated PAT RTRT│
 ├───────────────────────────────────┼───────────────────┼───────────────────┤
 │ Analytical Release Time           │ 14 days           │ Real-Time (Instant)│
 │ Material Quarantined              │ 500 kg API        │ 0 kg              │
 │ Sampling Frequency                │ 1 sample / batch  │ 360 scans / hour  │
 │ Off-Spec Material Produced        │ 45 kg (Whole batch)│ 1.2 kg (Diverted) │
 │ Quality Assurance Cost            │  $120 / sample │$ 0.50 / scan      │
 └───────────────────────────────────┴───────────────────┴───────────────────┘

# 6. Conclusions & Roadmap for Process Engineers

Implementing Continuous PAT and Real-Time Release Testing (RTRT) transforms pharmaceutical flow manufacturing:

Eliminates Release Testing Delays: API batches are released for packaging in real time upon completion of the flow campaign.
Prevents Batch Loss: Automated 3-way diverter valves reject transient out-of-spec fluid within milliseconds, protecting overall product quality.
Ensures Regulatory Compliance: Fulfills all ICH Q13 requirements for Level 1 and Level 2 process control strategies.

Continuous PAT & Real-Time Release Testing (RTRT) in Flow Skids: Sensor Fusion, Chemometrics & ICH Q13 Compliance

Peer-Reviewed & PE Verified

# Continuous PAT & Real-Time Release Testing (RTRT) in Flow Skids: Sensor Fusion, Chemometrics & ICH Q13 Compliance

# Executive Summary & Regulatory Context

# 1. Architectural Overview of Continuous PAT & Sensor Fusion

# 1.1 Comparison of Continuous PAT Technologies

# 2. Optical Physics & Analytical Cell Mechanics

# 2.1 Diamond ATR-FTIR Evanescent Wave Optics

# 2.2 Benchtop 60 MHz Quantitative NMR ( $q\text{NMR}$ )

# 3. Chemometrics & Multivariate Calibration Modeling

# 3.1 Spectral Preprocessing Protocols

# 3.2 Partial Least Squares (PLS) Regression Modeling

# 3.3 Anomaly Detection: Hotelling's $T^2$ & $Q$ -Residuals

# 4. ICH Q13 Regulatory Control Strategy & Diverter Valve Automation

# 4.1 Out-of-Spec (OOS) Material Diversion Dynamics

# 5. Industrial Case Study: RTRT Deployment on a Commercial $50\,\text{kg/day}$ API Skid

# 5.1 System Configuration & PAT Deployment

# 5.2 Results & Commercial Benefits

# 6. Conclusions & Roadmap for Process Engineers

Discussion

Peer-Reviewed & PE Verified

# Continuous PAT & Real-Time Release Testing (RTRT) in Flow Skids: Sensor Fusion, Chemometrics & ICH Q13 Compliance

# Executive Summary & Regulatory Context

# 1. Architectural Overview of Continuous PAT & Sensor Fusion

# 1.1 Comparison of Continuous PAT Technologies

# 2. Optical Physics & Analytical Cell Mechanics

# 2.1 Diamond ATR-FTIR Evanescent Wave Optics

# 2.2 Benchtop 60 MHz Quantitative NMR (qNMRq\text{NMR}qNMR)

# 3. Chemometrics & Multivariate Calibration Modeling

# 3.1 Spectral Preprocessing Protocols

# 3.2 Partial Least Squares (PLS) Regression Modeling

# 3.3 Anomaly Detection: Hotelling's T2T^2T2 & QQQ-Residuals

# 4. ICH Q13 Regulatory Control Strategy & Diverter Valve Automation

# 4.1 Out-of-Spec (OOS) Material Diversion Dynamics

# 5. Industrial Case Study: RTRT Deployment on a Commercial 50 kg/day50\,\text{kg/day}50kg/day API Skid

# 5.1 System Configuration & PAT Deployment

# 5.2 Results & Commercial Benefits

# 6. Conclusions & Roadmap for Process Engineers

Discussion

# 2.2 Benchtop 60 MHz Quantitative NMR ( $q\text{NMR}$ )

# 3.3 Anomaly Detection: Hotelling's $T^2$ & $Q$ -Residuals

# 5. Industrial Case Study: RTRT Deployment on a Commercial $50\,\text{kg/day}$ API Skid