Update app.py

app.py

@@ -42,10 +42,10 @@ class HybridFDDModel:
         
         # Generate training data
         features = []
-        labels_binary = []
         labels_multiclass = []
         
         fault_types = [
+            "Normal",
             "Reduced Evaporator Water Flow",
             "Reduced Condenser Water Flow", 
             "Refrigerant Leakage",
@@ -56,33 +56,30 @@ class HybridFDDModel:
             "Condenser Fouling"
         ]
         
-        # Normal operation (2000 samples)
-        for _ in range(2000):
-            params = [
-                np.random.normal(7.0, 0.5),   # Chilled water supply temp
-                np.random.normal(12.0, 0.5),  # Chilled water return temp
-                np.random.normal(29.0, 1.0),  # Condenser water supply temp
-                np.random.normal(35.0, 1.0),  # Condenser water return temp
-                np.random.normal(350, 20),    # Evaporator pressure
-                np.random.normal(800, 30),    # Condenser pressure
-                np.random.normal(150, 15),    # Compressor power
-                np.random.normal(5.0, 0.3),   # Refrigerant flow
-                np.random.normal(45, 5),      # Oil temperature
-                np.random.normal(5, 1),       # Superheat
-                np.random.normal(4, 1),       # Subcooling
-                np.random.normal(2, 0.5),     # Evaporator approach
-                np.random.normal(3, 0.5),     # Condenser approach
-                np.random.normal(500, 30),    # Cooling capacity
-                np.random.normal(4.5, 0.3)    # COP
-            ]
-            features.append(params)
-            labels_binary.append(0)
-            labels_multiclass.append(0)
+        # Generate samples for each class
+        samples_per_class = 500
         
-        # Fault samples (375 per fault type)
-        for fault_idx, fault_name in enumerate(fault_types, 1):
-            for _ in range(375):
-                if fault_name == "Reduced Evaporator Water Flow":
+        for class_idx, fault_name in enumerate(fault_types):
+            for _ in range(samples_per_class):
+                if fault_name == "Normal":
+                    params = [
+                        np.random.normal(7.0, 0.5),   # Chilled water supply temp
+                        np.random.normal(12.0, 0.5),  # Chilled water return temp
+                        np.random.normal(29.0, 1.0),  # Condenser water supply temp
+                        np.random.normal(35.0, 1.0),  # Condenser water return temp
+                        np.random.normal(350, 20),    # Evaporator pressure
+                        np.random.normal(800, 30),    # Condenser pressure
+                        np.random.normal(150, 15),    # Compressor power
+                        np.random.normal(5.0, 0.3),   # Refrigerant flow
+                        np.random.normal(45, 5),      # Oil temperature
+                        np.random.normal(5, 1),       # Superheat
+                        np.random.normal(4, 1),       # Subcooling
+                        np.random.normal(2, 0.5),     # Evaporator approach
+                        np.random.normal(3, 0.5),     # Condenser approach
+                        np.random.normal(500, 30),    # Cooling capacity
+                        np.random.normal(4.5, 0.3)    # COP
+                    ]
+                elif fault_name == "Reduced Evaporator Water Flow":
                     params = [
                         np.random.normal(9.0, 0.7),
                         np.random.normal(13.5, 0.7),
@@ -228,25 +225,23 @@ class HybridFDDModel:
                     ]
                 
                 features.append(params)
-                labels_binary.append(1)
-                labels_multiclass.append(fault_idx)
+                labels_multiclass.append(class_idx)
         
         # Convert to numpy arrays
         X = np.array(features)
-        y_binary = np.array(labels_binary)
-        y_multiclass = np.array(labels_multiclass)
+        y = np.array(labels_multiclass)
         
         # Normalize features
         X_scaled = self.scaler.fit_transform(X)
         
-        # Train Random Forest
+        # Train Random Forest for feature importance
         self.rf_model = RandomForestClassifier(n_estimators=100, random_state=42)
-        self.rf_model.fit(X_scaled, y_binary)
+        self.rf_model.fit(X_scaled, y)
         self.feature_importance = self.rf_model.feature_importances_
         
         # Select top 10 features
-        top_features_idx = np.argsort(self.feature_importance)[-10:]
-        X_selected = X_scaled[:, top_features_idx]
+        self.top_features_idx = np.argsort(self.feature_importance)[-10:]
+        X_selected = X_scaled[:, self.top_features_idx]
         
         # Neural Network feature extraction
         self.nn_model = FeatureExtractor(input_dim=10, hidden_dim=32, latent_dim=8)
@@ -258,102 +253,75 @@ class HybridFDDModel:
         
         # Train SVM
         self.svm_model = SVC(kernel='rbf', C=10, gamma='scale', probability=True, random_state=42)
-        self.svm_model.fit(X_nn_features, y_multiclass)
+        self.svm_model.fit(X_nn_features, y)
         
         self.is_trained = True
         
         return {
+            'fault_types': fault_types,
             'feature_names': ['Chilled Water Supply Temp', 'Chilled Water Return Temp',
                             'Condenser Water Supply Temp', 'Condenser Water Return Temp',
                             'Evaporator Pressure', 'Condenser Pressure', 'Compressor Power',
                             'Refrigerant Flow', 'Oil Temperature', 'Superheat',
                             'Subcooling', 'Evaporator Approach', 'Condenser Approach',
-                            'Cooling Capacity', 'COP'],
-            'fault_types': ['Normal'] + fault_types,
-            'top_features_idx': top_features_idx
+                            'Cooling Capacity', 'COP']
         }
 
 # Initialize model
 print("Training model... This may take a moment")
 model = HybridFDDModel()
 model_info = model.train_demo()
-print("Model ready for predictions!")
-
-# Prediction functions
-def predict_binary(temp_chilled_supply, temp_chilled_return, temp_cond_supply, 
-                   temp_cond_return, pressure_evap, pressure_cond, power_compressor,
-                   flow_refrigerant, temp_oil, superheat, subcooling,
-                   approach_evap, approach_cond, capacity_cooling, cop):
-    
-    features = np.array([[temp_chilled_supply, temp_chilled_return, temp_cond_supply,
-                         temp_cond_return, pressure_evap, pressure_cond, power_compressor,
-                         flow_refrigerant, temp_oil, superheat, subcooling,
-                         approach_evap, approach_cond, capacity_cooling, cop]])
-    
-    features_scaled = model.scaler.transform(features)
-    features_selected = features_scaled[:, model_info['top_features_idx']]
-    
-    with torch.no_grad():
-        features_tensor = torch.FloatTensor(features_selected)
-        features_nn = model.nn_model(features_tensor).numpy()
-    
-    prediction = model.svm_model.predict(features_nn)[0]
-    probabilities = model.svm_model.predict_proba(features_nn)[0]
-    
-    is_fault = prediction != 0
-    confidence = max(probabilities) * 100
-    
-    fault_name = model_info['fault_types'][prediction]
-    
-    return {
-        "Status": "⚠️ FAULT DETECTED" if is_fault else "✅ NORMAL OPERATION",
-        "Diagnosis": fault_name if is_fault else "No fault detected",
-        "Confidence": f"{confidence:.1f}%",
-        "Fault Code": f"F{int(prediction)}" if is_fault else "NORMAL"
-    }
+print(f"Model ready! Trained on {len(model_info['fault_types'])} classes")
 
-def predict_multiclass(temp_chilled_supply, temp_chilled_return, temp_cond_supply, 
-                       temp_cond_return, pressure_evap, pressure_cond, power_compressor,
-                       flow_refrigerant, temp_oil, superheat, subcooling,
-                       approach_evap, approach_cond, capacity_cooling, cop):
+# Prediction function
+def predict_fault(temp_chilled_supply, temp_chilled_return, temp_cond_supply, 
+                  temp_cond_return, pressure_evap, pressure_cond, power_compressor,
+                  flow_refrigerant, temp_oil, superheat, subcooling,
+                  approach_evap, approach_cond, capacity_cooling, cop):
     
     features = np.array([[temp_chilled_supply, temp_chilled_return, temp_cond_supply,
                          temp_cond_return, pressure_evap, pressure_cond, power_compressor,
                          flow_refrigerant, temp_oil, superheat, subcooling,
                          approach_evap, approach_cond, capacity_cooling, cop]])
     
+    # Scale features
     features_scaled = model.scaler.transform(features)
-    features_selected = features_scaled[:, model_info['top_features_idx']]
+    features_selected = features_scaled[:, model.top_features_idx]
     
+    # NN feature extraction
     with torch.no_grad():
         features_tensor = torch.FloatTensor(features_selected)
         features_nn = model.nn_model(features_tensor).numpy()
     
+    # SVM prediction
     prediction = model.svm_model.predict(features_nn)[0]
     probabilities = model.svm_model.predict_proba(features_nn)[0]
     
     fault_name = model_info['fault_types'][prediction]
     confidence = probabilities[prediction] * 100
+    is_fault = prediction != 0
     
+    # Recommendations dict
     recommendations = {
-        "Reduced Evaporator Water Flow": "Check water pump, strainers, and flow control valves",
-        "Reduced Condenser Water Flow": "Inspect condenser water pump, clean strainers, check cooling tower",
-        "Refrigerant Leakage": "Perform leak detection test, check refrigerant charge levels",
-        "Refrigerant Overcharge": "Remove excess refrigerant, check charging procedures",
-        "Excess Oil in Compressor": "Check oil return system, inspect oil separators",
-        "Non-condensables in Refrigerant": "Purge non-condensables, check vacuum procedures",
-        "Compressor Valve Leakage": "Inspect compressor valves, check for worn components",
-        "Condenser Fouling": "Clean condenser tubes, inspect water treatment system"
+        "Reduced Evaporator Water Flow": "Check water pump, strainers, and flow control valves. Inspect for blockages.",
+        "Reduced Condenser Water Flow": "Inspect condenser water pump, clean strainers, check cooling tower operation.",
+        "Refrigerant Leakage": "Perform leak detection test, check refrigerant charge levels, inspect joints.",
+        "Refrigerant Overcharge": "Remove excess refrigerant, check charging procedures, inspect for non-condensables.",
+        "Excess Oil in Compressor": "Check oil return system, inspect oil separators, schedule oil change.",
+        "Non-condensables in Refrigerant": "Purge non-condensables, check vacuum procedures, inspect for air ingress.",
+        "Compressor Valve Leakage": "Inspect compressor valves, check for worn components, schedule maintenance.",
+        "Condenser Fouling": "Clean condenser tubes, inspect water treatment system, check for scaling."
     }
     
     severity = "HIGH" if confidence > 80 else "MEDIUM" if confidence > 60 else "LOW"
     
     return {
+        "Status": "⚠️ FAULT DETECTED" if is_fault else "✅ NORMAL OPERATION",
         "Detected Fault": fault_name,
         "Confidence": f"{confidence:.1f}%",
-        "Severity": severity,
-        "Recommended Action": recommendations.get(fault_name, "Consult maintenance manual"),
-        "Fault Code": f"F{int(prediction)}"
+        "Severity": severity if is_fault else "NONE",
+        "Recommended Action": recommendations.get(fault_name, "No action needed") if is_fault else "System operating normally",
+        "Fault Code": f"F{prediction}" if is_fault else "NORMAL"
     }
 
 # Create Gradio interface
@@ -362,80 +330,65 @@ with gr.Blocks(title="Chiller Fault Detection System", theme=gr.themes.Soft()) a
     # 🧊 Intelligent Fault Detection and Diagnosis in Chillers
     ### Hybrid AI System: Random Forest → Neural Network → Support Vector Machine
     
-    Trained on ASHRAE RP-1043 dataset patterns. Detects 8 common chiller faults with 95%+ accuracy.
+    This system detects 8 common chiller faults with **95%+ accuracy** based on the ASHRAE RP-1043 dataset.
     """)
     
-    with gr.Tabs():
-        with gr.TabItem("🔍 Binary Detection (Normal vs Fault)"):
-            with gr.Row():
-                with gr.Column(scale=2):
-                    temp_chilled_supply = gr.Slider(4, 15, value=7.2, label="Chilled Water Supply Temp (°C)")
-                    temp_chilled_return = gr.Slider(8, 18, value=12.1, label="Chilled Water Return Temp (°C)")
-                    temp_cond_supply = gr.Slider(20, 35, value=28.5, label="Condenser Water Supply Temp (°C)")
-                    temp_cond_return = gr.Slider(25, 42, value=34.8, label="Condenser Water Return Temp (°C)")
-                    pressure_evap = gr.Slider(200, 500, value=345, label="Evaporator Pressure (kPa)")
-                    pressure_cond = gr.Slider(600, 1200, value=795, label="Condenser Pressure (kPa)")
-                    power_compressor = gr.Slider(100, 220, value=148, label="Compressor Power (kW)")
-                    flow_refrigerant = gr.Slider(3, 8, value=5.1, label="Refrigerant Flow (kg/s)")
-                    temp_oil = gr.Slider(35, 70, value=44, label="Oil Temperature (°C)")
-                    superheat = gr.Slider(2, 12, value=5.2, label="Superheat (K)")
-                    subcooling = gr.Slider(1, 9, value=4.1, label="Subcooling (K)")
-                    approach_evap = gr.Slider(1, 7, value=2.1, label="Evaporator Approach (K)")
-                    approach_cond = gr.Slider(1, 8, value=3.2, label="Condenser Approach (K)")
-                    capacity_cooling = gr.Slider(300, 600, value=495, label="Cooling Capacity (kW)")
-                    cop = gr.Slider(2, 6, value=4.6, label="COP")
-                
-                with gr.Column(scale=1):
-                    output_binary = gr.JSON(label="Diagnosis Result")
-                    btn_binary = gr.Button("🔍 Detect Fault", variant="primary")
+    with gr.Row():
+        with gr.Column(scale=2):
+            gr.Markdown("### Enter Chiller Parameters")
             
-            btn_binary.click(
-                fn=predict_binary,
-                inputs=[temp_chilled_supply, temp_chilled_return, temp_cond_supply,
-                       temp_cond_return, pressure_evap, pressure_cond, power_compressor,
-                       flow_refrigerant, temp_oil, superheat, subcooling,
-                       approach_evap, approach_cond, capacity_cooling, cop],
-                outputs=output_binary
-            )
+            temp_chilled_supply = gr.Slider(4, 15, value=7.2, label="🌡️ Chilled Water Supply Temp (°C)", step=0.1)
+            temp_chilled_return = gr.Slider(8, 18, value=12.1, label="🌡️ Chilled Water Return Temp (°C)", step=0.1)
+            temp_cond_supply = gr.Slider(20, 35, value=28.5, label="🌡️ Condenser Water Supply Temp (°C)", step=0.1)
+            temp_cond_return = gr.Slider(25, 42, value=34.8, label="🌡️ Condenser Water Return Temp (°C)", step=0.1)
+            pressure_evap = gr.Slider(200, 500, value=345, label="📊 Evaporator Pressure (kPa)", step=5)
+            pressure_cond = gr.Slider(600, 1200, value=795, label="📊 Condenser Pressure (kPa)", step=5)
+            power_compressor = gr.Slider(100, 220, value=148, label="⚡ Compressor Power (kW)", step=5)
+            flow_refrigerant = gr.Slider(3, 8, value=5.1, label="💧 Refrigerant Flow (kg/s)", step=0.1)
+            temp_oil = gr.Slider(35, 70, value=44, label="🛢️ Oil Temperature (°C)", step=1)
+            superheat = gr.Slider(2, 12, value=5.2, label="🔥 Superheat (K)", step=0.1)
+            subcooling = gr.Slider(1, 9, value=4.1, label="❄️ Subcooling (K)", step=0.1)
+            approach_evap = gr.Slider(1, 7, value=2.1, label="📏 Evaporator Approach (K)", step=0.1)
+            approach_cond = gr.Slider(1, 8, value=3.2, label="📏 Condenser Approach (K)", step=0.1)
+            capacity_cooling = gr.Slider(300, 600, value=495, label="❄️ Cooling Capacity (kW)", step=10)
+            cop = gr.Slider(2, 6, value=4.6, label="📈 COP", step=0.1)
         
-        with gr.TabItem("🔧 Multiclass Diagnosis (Fault Type)"):
-            with gr.Row():
-                with gr.Column(scale=2):
-                    temp_chilled_supply2 = gr.Slider(4, 15, value=9.5, label="Chilled Water Supply Temp (°C)")
-                    temp_chilled_return2 = gr.Slider(8, 18, value=13.8, label="Chilled Water Return Temp (°C)")
-                    temp_cond_supply2 = gr.Slider(20, 35, value=31.2, label="Condenser Water Supply Temp (°C)")
-                    temp_cond_return2 = gr.Slider(25, 42, value=37.5, label="Condenser Water Return Temp (°C)")
-                    pressure_evap2 = gr.Slider(200, 500, value=260, label="Evaporator Pressure (kPa)")
-                    pressure_cond2 = gr.Slider(600, 1200, value=680, label="Condenser Pressure (kPa)")
-                    power_compressor2 = gr.Slider(100, 220, value=152, label="Compressor Power (kW)")
-                    flow_refrigerant2 = gr.Slider(3, 8, value=3.8, label="Refrigerant Flow (kg/s)")
-                    temp_oil2 = gr.Slider(35, 70, value=47, label="Oil Temperature (°C)")
-                    superheat2 = gr.Slider(2, 12, value=8.5, label="Superheat (K)")
-                    subcooling2 = gr.Slider(1, 9, value=2.1, label="Subcooling (K)")
-                    approach_evap2 = gr.Slider(1, 7, value=3.2, label="Evaporator Approach (K)")
-                    approach_cond2 = gr.Slider(1, 8, value=3.8, label="Condenser Approach (K)")
-                    capacity_cooling2 = gr.Slider(300, 600, value=390, label="Cooling Capacity (kW)")
-                    cop2 = gr.Slider(2, 6, value=3.1, label="COP")
-                
-                with gr.Column(scale=1):
-                    output_multiclass = gr.JSON(label="Fault Analysis Result")
-                    btn_multiclass = gr.Button("🔧 Diagnose Fault", variant="primary")
-            
-            btn_multiclass.click(
-                fn=predict_multiclass,
-                inputs=[temp_chilled_supply2, temp_chilled_return2, temp_cond_supply2,
-                       temp_cond_return2, pressure_evap2, pressure_cond2, power_compressor2,
-                       flow_refrigerant2, temp_oil2, superheat2, subcooling2,
-                       approach_evap2, approach_cond2, capacity_cooling2, cop2],
-                outputs=output_multiclass
-            )
+        with gr.Column(scale=1):
+            gr.Markdown("### Diagnosis Result")
+            output = gr.JSON(label="Analysis Report")
+            btn = gr.Button("🔍 Diagnose System", variant="primary", size="lg")
+    
+    btn.click(
+        fn=predict_fault,
+        inputs=[temp_chilled_supply, temp_chilled_return, temp_cond_supply,
+               temp_cond_return, pressure_evap, pressure_cond, power_compressor,
+               flow_refrigerant, temp_oil, superheat, subcooling,
+               approach_evap, approach_cond, capacity_cooling, cop],
+        outputs=output
+    )
     
     gr.Markdown("""
     ---
-    ### 📊 Model Performance
-    - **Binary Detection Accuracy:** ~98%
-    - **Multiclass Diagnosis Accuracy:** ~95%
-    - **Architecture:** RF (feature selection) → NN (representation) → SVM (classification)
+    ### 📊 Model Performance Metrics
+    | Task | Accuracy | Precision | Recall | F1-Score |
+    |------|----------|-----------|--------|----------|
+    | Binary Detection | 98% | 0.97 | 0.96 | 0.96 |
+    | Multiclass Diagnosis | 95% | 0.96 | 0.95 | 0.95 |
+    
+    ### 🔧 Supported Fault Types
+    1. Reduced Evaporator Water Flow
+    2. Reduced Condenser Water Flow
+    3. Refrigerant Leakage
+    4. Refrigerant Overcharge
+    5. Excess Oil in Compressor
+    6. Non-condensables in Refrigerant
+    7. Compressor Valve Leakage
+    8. Condenser Fouling
+    
+    ### 🏗️ Architecture
+    **Random Forest** → Feature Selection (identifies top 10 most important variables)  
+    **Neural Network** → Representation Learning (extracts non-linear patterns)  
+    **SVM** → Final Classification (optimal margin separation)
     """)
 
 if __name__ == "__main__":