{ "cells": [ {"cell_type": "markdown", "metadata": {}, "source": [ "# \ud83c\udfae Roll Compactor Control Performance: PID Analysis & SPC\n", "\n", "**Dataset:** Roll Compactor Control Performance (Synthetic) v1.0 \n", "**Publisher:** [Innovative Process Applications (IPA)](https://www.innovativeprocess.com) \n", "**License:** CC BY 4.0 \n", "**Scientific basis:** Szappanos-Csord\u00e1s (2018), Chapter 3.1\n", "\n", "> \u26a0\ufe0f **Synthetic educational data** \u2014 not real measurements.\n", "\n", "---\n", "\n", "This notebook covers:\n", "1. Load both summary and time-series data\n", "2. Visualize PID step responses across control architectures\n", "3. Compare control quality metrics (CV%, settling time, overshoot)\n", "4. Build SPC control charts\n", "5. Quantify the twin feed screw advantage\n", "6. Classify control quality from time-series features" ]}, {"cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "import pandas as pd\n", "import numpy as np\n", "import matplotlib.pyplot as plt\n", "import seaborn as sns\n", "from sklearn.ensemble import RandomForestClassifier\n", "from sklearn.preprocessing import LabelEncoder\n", "from sklearn.model_selection import cross_val_score\n", "\n", "sns.set_style('whitegrid')\n", "plt.rcParams['figure.dpi'] = 100\n", "\n", "# Load data\n", "df_summary = pd.read_csv('control_performance_summary_v1.0.csv')\n", "df_ts = pd.read_csv('control_performance_timeseries_v1.0.csv')\n", "\n", "print(f'Summary: {df_summary.shape[0]} runs, {df_summary.shape[1]} columns')\n", "print(f'Time series: {df_ts.shape[0]} rows, {df_ts.shape[1]} columns')\n", "print(f'Control architectures: {df_summary[\"control_architecture\"].unique()}')\n", "df_summary.head()" ]}, {"cell_type": "markdown", "metadata": {}, "source": [ "## Part 1: PID Step Response Visualization\n", "\n", "The defining characteristic of a PID controller is its step response \u2014 how\n", "the actual value tracks a setpoint change. We\u2019ll compare the four control\n", "architectures responding to the same SCF step-up scenario." ]}, {"cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Pick one material and one scenario to compare architectures\n", "scenario = 'scf_step_up'\n", "material = 'MCC_Mannitol_Mix'\n", "\n", "architectures = ['no_gap_control', 'pid_gw_screw', 'pid_scf_gw', 'pid_scf_gw_twin_screw']\n", "labels = ['No Gap Control', 'PID: GW+Screw', 'PID: SCF+GW', 'PID: SCF+GW+Twin Screw']\n", "colors = ['#cc4444', '#d4a017', '#2E86C1', '#008080']\n", "\n", "fig, axes = plt.subplots(2, 1, figsize=(14, 8), sharex=True)\n", "\n", "for arch, label, color in zip(architectures, labels, colors):\n", " run = df_summary[(df_summary['control_architecture']==arch) &\n", " (df_summary['material']==material) &\n", " (df_summary['scenario']==scenario)]\n", " if len(run) == 0:\n", " continue\n", " rid = run.iloc[0]['run_id']\n", " ts = df_ts[df_ts['run_id']==rid]\n", "\n", " axes[0].plot(ts['time_s'], ts['scf_actual_kN_per_cm'], label=label,\n", " color=color, alpha=0.8, linewidth=1.2)\n", " axes[1].plot(ts['time_s'], ts['gw_actual_mm'], label=label,\n", " color=color, alpha=0.8, linewidth=1.2)\n", "\n", "# Setpoint lines\n", "axes[0].axhline(4.0, color='gray', linestyle=':', alpha=0.5)\n", "axes[0].axhline(8.0, color='gray', linestyle=':', alpha=0.5)\n", "axes[0].axvline(30, color='black', linestyle='--', alpha=0.3, label='Setpoint change')\n", "axes[0].set_ylabel('SCF (kN/cm)')\n", "axes[0].set_title(f'PID Step Response Comparison \\u2014 SCF Step Up (4\\u21928 kN/cm), {material}')\n", "axes[0].legend(loc='lower right', fontsize=9)\n", "\n", "axes[1].axvline(30, color='black', linestyle='--', alpha=0.3)\n", "axes[1].set_ylabel('Gap Width (mm)')\n", "axes[1].set_xlabel('Time (s)')\n", "axes[1].set_title('Gap Width Response During SCF Step Change')\n", "axes[1].legend(loc='upper right', fontsize=9)\n", "\n", "plt.tight_layout()\n", "plt.show()\n", "\n", "print('Key observations:')\n", "print('- No gap control: slow approach, steady-state offset, noisy')\n", "print('- PID SCF+GW: overshoot then clean settling')\n", "print('- Twin screw: tightest band, fastest settling, least noise')" ]}, {"cell_type": "markdown", "metadata": {}, "source": [ "## Part 2: Control Quality Metrics by Architecture\n", "\n", "CV% (coefficient of variation) is the primary metric from Chapter 3.1 of\n", "the dissertation. Lower CV = more robust process." ]}, {"cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "fig, axes = plt.subplots(1, 3, figsize=(16, 4))\n", "\n", "order = ['no_gap_control', 'pid_gw_screw', 'pid_scf_gw', 'pid_scf_gw_twin_screw']\n", "palette = ['#cc4444', '#d4a017', '#2E86C1', '#008080']\n", "\n", "sns.boxplot(data=df_summary, x='control_architecture', y='scf_ss_cv_pct',\n", " order=order, palette=palette, ax=axes[0])\n", "axes[0].set_title('SCF Steady-State CV%')\n", "axes[0].set_ylabel('CV %')\n", "axes[0].tick_params(axis='x', rotation=30)\n", "axes[0].set_xlabel('')\n", "\n", "sns.boxplot(data=df_summary, x='control_architecture', y='gw_ss_cv_pct',\n", " order=order, palette=palette, ax=axes[1])\n", "axes[1].set_title('Gap Width Steady-State CV%')\n", "axes[1].set_ylabel('CV %')\n", "axes[1].tick_params(axis='x', rotation=30)\n", "axes[1].set_xlabel('')\n", "\n", "sns.boxplot(data=df_summary, x='control_architecture', y='scf_settling_time_s',\n", " order=order, palette=palette, ax=axes[2])\n", "axes[2].set_title('SCF Settling Time')\n", "axes[2].set_ylabel('Time (s)')\n", "axes[2].tick_params(axis='x', rotation=30)\n", "axes[2].set_xlabel('')\n", "\n", "plt.tight_layout()\n", "plt.show()\n", "\n", "print('\\nMean CV% by architecture:')\n", "print(df_summary.groupby('control_architecture')[['scf_ss_cv_pct','gw_ss_cv_pct']].mean().round(2).loc[order])" ]}, {"cell_type": "markdown", "metadata": {}, "source": [ "## Part 3: Material Effect on Controllability\n", "\n", "Brittle materials (mannitol) produce more erratic force signals due to\n", "particle fragmentation, making the control task harder." ]}, {"cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "fig, axes = plt.subplots(1, 2, figsize=(12, 4))\n", "mat_order = ['MCC_101', 'MCC_Mannitol_Mix', 'Mannitol_SD']\n", "mat_colors = ['#008080', '#2E86C1', '#d4a017']\n", "\n", "sns.boxplot(data=df_summary, x='material', y='scf_ss_cv_pct',\n", " order=mat_order, palette=mat_colors, ax=axes[0])\n", "axes[0].set_title('SCF CV% by Material')\n", "axes[0].set_ylabel('SCF CV %')\n", "axes[0].set_xlabel('')\n", "\n", "sns.boxplot(data=df_summary, x='material', y='gw_ss_cv_pct',\n", " order=mat_order, palette=mat_colors, ax=axes[1])\n", "axes[1].set_title('Gap Width CV% by Material')\n", "axes[1].set_ylabel('GW CV %')\n", "axes[1].set_xlabel('')\n", "\n", "plt.tight_layout()\n", "plt.show()\n", "\n", "print('Interpretation:')\n", "print('- MCC (plastic): smoothest compaction, easiest to control')\n", "print('- Mannitol (brittle): particle fragmentation creates force spikes')\n", "print('- Mixture: intermediate behavior')" ]}, {"cell_type": "markdown", "metadata": {}, "source": [ "## Part 4: SPC Control Charts\n", "\n", "Statistical Process Control charts are standard tools in pharma\n", "manufacturing. Let\u2019s build X-bar and R charts for a sample run." ]}, {"cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Pick the twin-screw + MCC steady-state baseline for the cleanest signal\n", "run_spc = df_summary[(df_summary['control_architecture']=='pid_scf_gw_twin_screw') &\n", " (df_summary['material']=='MCC_101') &\n", " (df_summary['scenario']=='steady_state_baseline')]\n", "if len(run_spc) > 0:\n", " rid = run_spc.iloc[0]['run_id']\n", " ts = df_ts[df_ts['run_id']==rid].copy()\n", "\n", " # Use only steady-state portion (t > 30s)\n", " ts_ss = ts[ts['time_s'] >= 30.0].copy()\n", "\n", " scf_mean = ts_ss['scf_actual_kN_per_cm'].mean()\n", " scf_std = ts_ss['scf_actual_kN_per_cm'].std()\n", " ucl = scf_mean + 3 * scf_std\n", " lcl = scf_mean - 3 * scf_std\n", "\n", " fig, ax = plt.subplots(figsize=(14, 4))\n", " ax.plot(ts_ss['time_s'], ts_ss['scf_actual_kN_per_cm'],\n", " 'o-', markersize=3, color='#008080', alpha=0.7)\n", " ax.axhline(scf_mean, color='#1B2A3B', linewidth=2, label=f'CL = {scf_mean:.3f}')\n", " ax.axhline(ucl, color='red', linestyle='--', label=f'UCL = {ucl:.3f}')\n", " ax.axhline(lcl, color='red', linestyle='--', label=f'LCL = {lcl:.3f}')\n", " ax.fill_between(ts_ss['time_s'], lcl, ucl, alpha=0.05, color='red')\n", " ax.set_xlabel('Time (s)')\n", " ax.set_ylabel('SCF (kN/cm)')\n", " ax.set_title(f'X-bar Control Chart \\u2014 Twin Screw PID, MCC 101, Steady State')\n", " ax.legend(loc='upper right')\n", " plt.tight_layout()\n", " plt.show()\n", "\n", " print(f'Process capability summary:')\n", " print(f' Mean: {scf_mean:.3f} kN/cm')\n", " print(f' Std: {scf_std:.4f} kN/cm')\n", " print(f' CV: {scf_std/scf_mean*100:.2f}%')\n", " spec_half = 0.5 # example spec: setpoint +/- 0.5 kN/cm\n", " cpk = min(ucl - scf_mean, scf_mean - lcl) / (3 * scf_std)\n", " print(f' Cpk (\\u00b13\\u03c3 natural): {cpk:.2f}')" ]}, {"cell_type": "markdown", "metadata": {}, "source": [ "## Part 5: Twin Feed Screw Advantage\n", "\n", "This is the key IPA design differentiator. Twin feed screws provide more\n", "uniform powder delivery, which directly reduces process variability." ]}, {"cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Compare single vs twin screw (both with SCF+GW PID)\n", "single = df_summary[df_summary['control_architecture']=='pid_scf_gw']\n", "twin = df_summary[df_summary['control_architecture']=='pid_scf_gw_twin_screw']\n", "\n", "metrics = ['scf_ss_cv_pct', 'gw_ss_cv_pct', 'scf_settling_time_s', 'scf_overshoot_pct']\n", "metric_labels = ['SCF CV%', 'GW CV%', 'Settling Time (s)', 'Overshoot %']\n", "\n", "fig, axes = plt.subplots(1, 4, figsize=(16, 4))\n", "for ax, metric, mlabel in zip(axes, metrics, metric_labels):\n", " data = pd.DataFrame({\n", " 'Single Screw': single[metric].values,\n", " 'Twin Screw': twin[metric].values,\n", " })\n", " data.plot.box(ax=ax, color=dict(boxes='#008080', whiskers='gray',\n", " medians='#1B2A3B', caps='gray'))\n", " ax.set_title(mlabel)\n", " ax.set_ylabel(mlabel)\n", "\n", "plt.suptitle('Single Screw vs. Twin Feed Screw (Same PID Architecture)',\n", " fontsize=13, y=1.02)\n", "plt.tight_layout()\n", "plt.show()\n", "\n", "print('\\nImprovement with twin screw:')\n", "for metric, mlabel in zip(metrics, metric_labels):\n", " s = single[metric].mean()\n", " t = twin[metric].mean()\n", " improvement = (s - t) / s * 100 if s > 0 else 0\n", " print(f' {mlabel}: {s:.2f} \\u2192 {t:.2f} ({improvement:+.1f}% improvement)')" ]}, {"cell_type": "markdown", "metadata": {}, "source": [ "## Part 6: Classify Control Quality from Metrics\n", "\n", "Can we predict the control quality grade from the summary metrics? This\n", "demonstrates how machine learning can support process monitoring." ]}, {"cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# Encode targets and features\n", "le = LabelEncoder()\n", "y = le.fit_transform(df_summary['control_quality_grade'])\n", "\n", "feature_cols = ['scf_ss_cv_pct', 'gw_ss_cv_pct', 'scf_deviation_from_setpoint_pct',\n", " 'gw_deviation_from_setpoint_pct', 'scf_settling_time_s',\n", " 'gw_settling_time_s', 'scf_overshoot_pct']\n", "X = df_summary[feature_cols].values\n", "\n", "rf = RandomForestClassifier(n_estimators=100, random_state=42)\n", "scores = cross_val_score(rf, X, y, cv=5, scoring='accuracy')\n", "print(f'RF 5-fold CV accuracy: {scores.mean():.3f} \\u00b1 {scores.std():.3f}')\n", "\n", "rf.fit(X, y)\n", "importances = pd.Series(rf.feature_importances_, index=feature_cols).sort_values()\n", "\n", "fig, ax = plt.subplots(figsize=(8, 4))\n", "importances.plot.barh(color='#008080', ax=ax)\n", "ax.set_title('Feature Importance for Control Quality Grade Prediction')\n", "ax.set_xlabel('Importance')\n", "plt.tight_layout()\n", "plt.show()" ]}, {"cell_type": "markdown", "metadata": {}, "source": [ "## Takeaways\n", "\n", "1. **Control architecture matters enormously:** Going from no gap control to\n", " full PID SCF+GW control reduces process variability by 50\u201370%. This is\n", " not optional in regulated pharma manufacturing.\n", "\n", "2. **Twin feed screws are a multiplicative advantage:** On top of PID control,\n", " twin screws provide an additional 20\u201330% reduction in CV by smoothing\n", " feed rate fluctuations. This is a mechanical design choice, not a tuning\n", " parameter \u2014 it must be built into the machine.\n", "\n", "3. **Material properties affect control difficulty:** Brittle materials\n", " produce noisier signals that challenge even the best PID controllers.\n", " Process development must account for this.\n", "\n", "4. **SPC tools apply directly to continuous granulation:** Control charts,\n", " Cp/Cpk, and trend analysis are standard pharma quality tools that\n", " connect naturally to roll compaction process data.\n", "\n", "---\n", "\n", "For twin-feed-screw roller compactors with advanced PID control and direct\n", "engineer support, see IPA\u2019s CL-series compactors:\n", "**https://www.innovativeprocess.com**\n", "\n", "*Dataset \u00a9 2026 Innovative Process Applications, CC BY 4.0.* \n", "*Scientific basis: Szappanos-Csord\u00e1s (2018), Chapter 3.1.*" ]} ], "metadata": { "colab": { "provenance": [], "toc_visible": true, "authorship_tag": "Innovative Process Applications (IPA)" }, "kernelspec": {"display_name": "Python 3", "language": "python", "name": "python3"}, "language_info": {"name": "python", "version": "3.11"} }, "nbformat": 4, "nbformat_minor": 5 }