omiran/predictive_maintenance

1# app.py2 3from flask import Flask, render_template, request4import pandas as pd5import torch6#from model import predict7import numpy as np8import torch.nn as nn9import torch.optim as optim10from torch.utils.data import DataLoader, TensorDataset11 12from flask import Flask, render_template, request, jsonify13import pandas as pd14import numpy as np15import torch16from sklearn.preprocessing import StandardScaler17if torch.cuda.is_available():18    device = torch.device("cuda")  # Use GPU19else:20    device = torch.device("cpu")   # Use CPU21 22app = Flask(__name__)23 24# Define the neural network architecture25class MultiOutputNN(nn.Module):26    def __init__(self, input_dim, output_dims):27        super(MultiOutputNN, self).__init__()28        self.shared_hidden_layer = nn.Sequential(29            nn.Linear(input_dim, 64),30            nn.ReLU()31        )32        self.output_layers = nn.ModuleList([33            nn.Linear(64, out_dim) for out_dim in output_dims34        ])35                36    def forward(self, x):37        shared_output = self.shared_hidden_layer(x)38        print(f'shared_output shape: {shared_output.shape}')39                                    40        # Before matrix multiplication41        print(f'input shape: {x.shape}')42        43        outputs = [output_layer(shared_output) for output_layer in self.output_layers]44        for i, output in enumerate(outputs):45            print(f'output {i} shape: {output.shape}')46        return outputs47 48 49 50# Define the upload folder51app.config['UPLOAD_FOLDER'] = 'uploads'52 53# Dummy preprocessing function54def preprocess_input(vibration_file, gas_file):55    vibration_data = pd.read_csv(vibration_file)56    gas_data = pd.read_csv(gas_file, sep=';')57 58    # Perform any necessary preprocessing here59    60    gas_data.drop(['Unnamed: 15', 'Unnamed: 16', 'Date', 'Time', 'NMHC(GT)'], axis=1, inplace=True)61    cleaned_gas_data = gas_data.dropna()62    print(cleaned_gas_data.info())63    def replace_comma_with_period_in_columns(df, columns):64        for column in columns:65            df[column] = df[column].str.replace(',', '.', regex=False)66        return df67 68    comma_col = ['CO(GT)', 'C6H6(GT)', 'T', 'RH', 'AH']69    cleaned_gas_data = replace_comma_with_period_in_columns(cleaned_gas_data, comma_col)70 71    for i in comma_col:72        cleaned_gas_data[i] = pd.to_numeric(cleaned_gas_data[i], errors='coerce')73 74    75    # Concatenate the data76    processed_data = np.hstack((vibration_data.iloc[:9357, :], cleaned_gas_data))77 78    # Standardize the data79    scaler = StandardScaler()80    processed_data_scaled = pd.DataFrame(scaler.fit_transform(processed_data))81    processed_data_scaled = processed_data_scaled.values82 83        # Convert processed data to tensor84    X_data_tensor = torch.Tensor(processed_data_scaled ).to(device)85 86    return X_data_tensor87 88 89# this function sends email to representatives if the abnormality is detected 90# in the machine from the vibration sensor or if a gas is detected in environment 91# from the gas sensor92 93import smtplib94from email.mime.text import MIMEText95from email.mime.multipart import MIMEMultipart96 97def send_email(subject, body):98    sender_email = 'bidehassan@gmail.com' 99    sender_password = 'rmih ytdp znow dgjw'100    recipient_email = 'bidehassan@gmail.com'101 102    message = MIMEMultipart()103    message['From'] = sender_email104    message['To'] = recipient_email105    message['Subject'] = subject106 107    message.attach(MIMEText(body, 'plain'))108 109    try:110        server = smtplib.SMTP('smtp.gmail.com', 587)111        server.starttls()112        server.login(sender_email, sender_password)113        server.sendmail(sender_email, recipient_email, message.as_string())114        server.quit()115        print("Email sent successfully")116    except Exception as e:117        print(f"Failed to send email. Error: {e}")118 119# Example usage:120# send_email("Anomaly Detected", "Anomalies have been detected in both gas and vibration sensors.")121 122 123# gas sensor detection function124def detect_gas_anomaly(gas_sensor):125    thresholds = {126    'CO(GT)': gas_sensor['CO(GT)'].mean() - 2 * gas_sensor['CO(GT)'].std(),127    'PT08.S1(CO)': gas_sensor['PT08.S1(CO)'].mean() - 2 * gas_sensor['PT08.S1(CO)'].std(),128    'C6H6(GT)': gas_sensor['C6H6(GT)'].mean() - 2 * gas_sensor['C6H6(GT)'].std(),129    'PT08.S2(NMHC)': gas_sensor['PT08.S2(NMHC)'].mean() - 2 * gas_sensor['PT08.S2(NMHC)'].std(),130    'NOx(GT)': gas_sensor['NOx(GT)'].mean() - 2 * gas_sensor['NOx(GT)'].std(),131    'PT08.S3(NOx)': gas_sensor['PT08.S3(NOx)'].mean() - 2 * gas_sensor['PT08.S3(NOx)'].std(),132    'NO2(GT)': gas_sensor['NO2(GT)'].mean() - 2 * gas_sensor['NO2(GT)'].std(),133    'PT08.S4(NO2)': gas_sensor['PT08.S4(NO2)'].mean() - 2 * gas_sensor['PT08.S4(NO2)'].std(),134    'PT08.S5(O3)': gas_sensor['PT08.S5(O3)'].mean() - 2 * gas_sensor['PT08.S5(O3)'].std(),135    'T': gas_sensor['T'].mean() - 2 * gas_sensor['T'].std(),136    'RH': gas_sensor['RH'].mean() - 2 * gas_sensor['RH'].std(),137    'AH': gas_sensor['AH'].mean() - 2 * gas_sensor['AH'].std()138}139    140    # # Create a DataFrame to store anomaly flags141    # anomalies = pd.DataFrame(index=gas_data.index)142    143    # for parameter in thresholds.keys():144    #     # Detect anomalies for each parameter145    #     is_anomaly = gas_data[parameter] < thresholds[parameter]146    #     anomalies[f'{parameter}_Anomaly'] = is_anomaly.astype(int)147    148    # return anomalies149    anomalies = []150 151    for _, data_point in gas_sensor.iterrows():152        data_point_anomaly = {}153        for parameter, threshold in thresholds.items():154            data_point_anomaly[f'{parameter}_Anomaly'] = 1 if data_point[parameter] < threshold else 0155        anomalies.append(data_point_anomaly)156 157        return anomalies158 159# vibration sensor abnormality detection function160# def detect_vibration_anomaly(vibration_data):161#     thresholds = {162#         'Vibration_1': 1.226e-1,163#         'Vibration_2': 2.413e-1,164#         'Vibration_3': 1.187e-1165#     }166    167#     # Create a DataFrame to store anomaly flags168#     anomalies = pd.DataFrame(index=range(len(vibration_data)))  # Assuming list of lists169    170#     for i, sensor_readings in enumerate(vibration_data):171#         for sensor, threshold in thresholds.items():172#             # Detect anomalies for each sensor173#             is_anomaly = sensor_readings[i] > threshold174#             anomalies[f'{sensor}_Anomaly'] = is_anomaly.astype(int)175    176#     return anomalies177 178# def detect_vibration_anomaly(vibration_data):179#     thresholds = {180#         'Vibration_1': 1.226e-1,181#         'Vibration_2': 2.413e-1,182#         'Vibration_3': 1.187e-1183#     }184    185    # Create a list to store anomaly flags186    anomalies = []187 188    # for sensor_readings in vibration_data:189    #     sensor_anomalies = {}  # Store anomalies for each sensor190    #     for i, (sensor, threshold) in enumerate(thresholds.items()):191    #         # Detect anomalies for each sensor192    #         is_anomaly = sensor_readings[i+2] > threshold  # Assuming sensor data starts from index 2193    #         sensor_anomalies[f'{sensor}_Anomaly'] = int(is_anomaly)194    #     anomalies.append(sensor_anomalies)195 196    # return anomalies197 198    # for data_point in vibration_data:199    #     data_point_anomaly = {}200    #     for i, (sensor, threshold) in enumerate(thresholds.items()):201    #         data_point_anomaly[f'{sensor}_Anomaly'] = 1 if data_point[i+2] > threshold else 0202    #     anomalies.append(data_point_anomaly)203    204    #     return anomalies205 206# def detect_vibration_anomaly(predictions):207#     threshold = -9.8  # Set your threshold value208 209#     # Create a list to store anomaly flags210#     anomalies = []211 212#     for prediction in predictions:213#         data_point_anomaly = {}214#         data_point_anomaly['Vibration_1_Anomaly'] = 1 if prediction < threshold else 0215#         data_point_anomaly['Vibration_2_Anomaly'] = 1 if prediction < threshold else 0216#         data_point_anomaly['Vibration_3_Anomaly'] = 1 if prediction < threshold else 0217#         anomalies.append(data_point_anomaly)218 219#         return anomalies220def detect_vibration_anomaly(vibration_data):221    predicted_values = vibration_data  # Replace with your actual predicted values222    223    # Calculate mean and standard deviation224    mean_value = np.mean(predicted_values)225    std_dev = np.std(predicted_values)226    227    # Define a multiplier (e.g., 2 for 2 standard deviations)228    multiplier = 2229    230    # Calculate threshold231    threshold = mean_value - (multiplier * std_dev)232    233    # Create a list to store anomaly flags234    anomalies = [1 if value < threshold else 0 for value in predicted_values]235    236    return anomalies237 238 239 240    # Use the model to make predictions241def predict(input_data):242    # Load the model243    model = MultiOutputNN(input_dim=17, output_dims=[1, 11])244    model.load_state_dict(torch.load('multi_output_model.pth'))245 246    model.eval()247    with torch.no_grad():248        outputs = model(input_data)249        print(outputs)250        regression_prediction = outputs[0].cpu().numpy() #item()  # Assuming first output is regression251        classification_prediction = outputs[1].cpu().numpy() #item()  # Assuming second output is classification252 253    return regression_prediction, classification_prediction254 255 256# Route for the home page257@app.route('/')258def home():259    return render_template('index.html')260 261# Route to handle file uploads262@app.route('/upload', methods=['POST'])263def upload_files():264    vibration_file = request.files['vibration_data']265    gas_file = request.files['gas_data']266 267    vibration_path = f"{app.config['UPLOAD_FOLDER']}/vibration.csv"268    print(vibration_path)269    gas_path = f"{app.config['UPLOAD_FOLDER']}/gas.csv"270 271    vibration_file.save(vibration_path)272    gas_file.save(gas_path)273 274    # Preprocess the uploaded files275    processed_input = preprocess_input(vibration_path, gas_path)276 277    # Use the `processed_input` in your predict function278    prediction = predict(processed_input) 279    regression_prediction, classification_prediction = predict(processed_input)  # Include regression prediction280 281    # Read and process the gas and vibration data282    gas_data = pd.read_csv(gas_path, sep=';')283    gas_data.drop(['Unnamed: 15', 'Unnamed: 16', 'Date', 'Time', 'NMHC(GT)'], axis=1, inplace=True)284 285    def replace_comma_with_period_in_columns(df, columns):286        for column in columns:287            df[column] = df[column].str.replace(',', '.', regex=False)288        return df289 290    comma_col = ['CO(GT)', 'C6H6(GT)', 'T', 'RH', 'AH']291    gas_data = replace_comma_with_period_in_columns(gas_data, comma_col)292 293    for i in comma_col:294        gas_data[i] = pd.to_numeric(gas_data[i], errors='coerce')295    296    gas_anomaly =  gas_data.copy() # have a copy of the dataframe before converted tp list297    gas_data = gas_data.values.tolist()298 299    vibration_data = pd.read_csv(vibration_path).values[:9357, :].tolist()300    vib_data = pd.read_csv(vibration_path).values[:9357, :]301 302     # Detect anomalies in vibration data303    anomalies_vibration = detect_vibration_anomaly(prediction[0])304 305    # Detect anomalies in gas data306    gas_sensor = pd.DataFrame(gas_data, columns=['CO(GT)', 'PT08.S1(CO)', 'C6H6(GT)', 'PT08.S2(NMHC)', 'NOx(GT)', 'PT08.S3(NOx)', 'NO2(GT)', 'PT08.S4(NO2)', 'PT08.S5(O3)', 'T', 'RH', 'AH'])307    anomalies_gas = detect_gas_anomaly(gas_sensor)308 309    # Send email if anomalies are detected310    if anomalies_vibration or anomalies_gas:311        send_email("Anomaly Detected", "Anomaly detected in the system!")312 313    return render_template('result.html', prediction=prediction, gas_data=gas_data, vibration_data=vibration_data, anomalies_vibration=anomalies_vibration, anomalies_gas=anomalies_gas)314 315if __name__ == '__main__':316    app.run(debug=True)317