Spaces:
Sleeping
Sleeping
| # Everything here goes into the app.py file | |
| # Define Gradio inference function | |
| # 15 March 2025 | |
| # Prakash Kota | |
| # East Greenbush | |
| import numpy as np | |
| import tensorflow as tf | |
| import os | |
| import gradio as gr | |
| import joblib | |
| # Fixed reactor parameters | |
| V = 100 # Reactor volume (L) | |
| k = 0.1 # Reaction rate constant (1/min) | |
| delta_H = -50000 # Heat of reaction (J/mol) | |
| rho = 1 # Density in kg/L | |
| Cp = 4184 # Heat capacity in J/kg·K | |
| # Define the model directory | |
| model_dir = "./model" | |
| # Define Gradio inference function | |
| def predict_cstr(CA_in, T_in, F): | |
| # Load scalers and model | |
| scaler_X = joblib.load(f"{model_dir}/scaler_X.pkl") | |
| scaler_Y = joblib.load(f"{model_dir}/scaler_Y.pkl") | |
| model = tf.keras.models.load_model(f"{model_dir}/cstr_model.keras") | |
| # Scale input | |
| input_scaled = scaler_X.transform([[CA_in, T_in, F]]) | |
| # Predict using the model | |
| prediction_scaled = model.predict(input_scaled) | |
| prediction_original = scaler_Y.inverse_transform(prediction_scaled) | |
| CA_ss_pred, T_ss_pred = prediction_original[0] | |
| # Compute analytical solution | |
| CA_ss_analytical = (CA_in * (F / V)) / ((F / V) + k) | |
| T_ss_analytical = T_in + ((-delta_H * k * CA_ss_analytical) / (rho * Cp)) * (V / F) | |
| # Compute % Error and % Accuracy for Concentration | |
| #percent_error_CA = abs((CA_ss_pred - CA_ss_analytical) / CA_ss_analytical) * 100 | |
| percent_accuracy_CA = (1 - abs(CA_ss_pred - CA_ss_analytical) / CA_ss_analytical) * 100 | |
| # Compute % Error and % Accuracy for Temperature | |
| #percent_error_T = abs((T_ss_pred - T_ss_analytical) / T_ss_analytical) * 100 | |
| percent_accuracy_T = (1 - abs(T_ss_pred - T_ss_analytical) / T_ss_analytical) * 100 | |
| return (f"Predicted CA_ss: {CA_ss_pred:.4f} mol/L\n" | |
| f"Predicted T_ss: {T_ss_pred:.2f} K\n" | |
| f"\n" | |
| f"Analytical CA_ss: {CA_ss_analytical:.4f} mol/L\n" | |
| f"Analytical T_ss: {T_ss_analytical:.2f} K\n" | |
| f"\n" | |
| f"Accuracy\n" | |
| f"% Concentration Accuracy: {percent_accuracy_CA:.2f}%\n" | |
| f"% Temperature Accuracy: {percent_accuracy_T:.2f}%\n") | |
| # Deploy using Gradio | |
| iface = gr.Interface( | |
| fn=predict_cstr, | |
| inputs=[ | |
| gr.Number(label="Input Concentration CA_in - Range [0.5-2.0] mol/L"), | |
| gr.Number(label="Input Temperature T_in - Range [300-350] K"), | |
| gr.Number(label="CSTR Flow Rate F - Range [5-20] L/min") | |
| ], | |
| outputs="text", | |
| title="CSTR Surrogate Model Inference", | |
| description="Enter the input values to predict steady-state concentration and temperature." | |
| ) | |
| # Add the Markdown footer with a clickable hyperlink | |
| footer = gr.Markdown( | |
| 'For details about the model, please see the article - ' | |
| '[Bringing Historical Process Data to Life: Unlocking AI’s Goldmine with Neural Networks for Smarter Manufacturing](https://prakashkota.com/2025/03/16/bringing-historical-process-data-to-life-unlocking-ais-goldmine-with-neural-networks-for-smarter-manufacturing/)' | |
| ) | |
| # Launch the interface with the footer | |
| with gr.Blocks() as demo: | |
| iface.render() | |
| footer.render() | |
| # Ensure the app launches when executed | |
| if __name__ == "__main__": | |
| demo.launch(share=True) |