⚙️ PREDICTIVE MAINTENANCE PROJECT

Bearing Fault Detection and Classification

Predictive Maintenance with FFT Signal Processing and Random Forest Machine Learning

%100

accuracy

12kHz

sampling

FFT

Signal Analysis

R.F.

Random Forest

🎯

1.1 General Purpose of the Project

The running parts (bearings, transmissions, etc.) of vehicles used in the defense industry and heavy industry conditions need to be analyzed in terms of sustainability. Using sensor data instead of traditional maintenance methods condition-based predictive maintenance (Predictive Maintenance) is planned to be done.

📊 Data Source

Case Western Reserve University (CWRU) laboratory data

🎯 Target

Pre-detection and classification of the fault

📈

2.1 Signal Processing and Frequency Analysis Phase

During the analysis phase, data obtained from the CWRU website 12k Drive End data was used. For analysis and processing of signals Python programming language was used.

Data Loading

Uploading data to the system and drawing time-dependent graphs.

Time Series Analysis

Defective bearing blows It was observed that .

FFT (Fast Fourier Transform)

FFT process was applied to the signals to find the frequency of the pulses. In case of faulty signal around 160 Hz An increase in energy was detected.

✅ Result: It has been seen that these frequencies coincide with theoretical calculations.

💻 Signal Processing Code

# Source - https://stackoverflow.com/q (Modified by Community, CC BY-SA 4.0)

import scipy.io
import matplotlib.pyplot as plt
import numpy as np
from scipy.fft import fft, fftfreq

# Dataları yükleme kısmı
mat = scipy.io.loadmat('arizali.mat')
sinyal = mat['X105_DE_time'].reshape(-1)

# Perform FFT with SciPy
N = len(sinyal)
fs = 12000
yf = fft(sinyal)
xf = fftfreq(N, 1/fs)

# Pozitif frekansları alma
idx_max = N // 2

# Grafik Çizimi
plt.figure(figsize=(10, 6))
plt.plot(xf[:idx_max], np.abs(yf[:idx_max]))
plt.title('Arızalı Rulman Frekans Spektrumu')
plt.grid()
plt.show()

📊Figure 1 - Time Series Analysis | Figure 2 - Frequency Spectrum

🔬

3.1 Feature Extraction and Artificial Intelligence Application

Since direct analysis of raw signals is difficult, feature extraction It was decided to do so. signals into pieces of 1000 The following attributes were calculated for each part:

attribute	formula	Description
RMS	`√(mean(x²))`	Effective value of the signal
kurtosis	`scipy.stats.kurtosis()`	The sharpness measure of the distribution
Max Value	`max(\|x\|)`	Maximum absolute value

📊 Observation: It is seen in the graph that healthy and faulty data are separated from each other when calculated based on RMS values.

💻 Feature Extraction Code

# Source - StackOverflow (CC BY-SA 4.0)
import pandas as pd
from scipy.stats import kurtosis

def ozellik_cikar(sinyal, etiket):
    ozellikler = []
    parca_boyutu = 1000
    for i in range(0, len(sinyal) - parca_boyutu, parca_boyutu):
        parca = sinyal[i : i + parca_boyutu]
        
        # RMS ve diğer özellikler hesaplanıyor
        rms = np.sqrt(np.mean(parca**2))
        kurt = kurtosis(parca)
        max_val = np.max(np.abs(parca))
        
        ozellikler.append([rms, max_val, kurt, etiket])
    return ozellikler

# DataFrame oluşturma
df_features = pd.DataFrame(ozellikler, 
    columns=['RMS', 'Max_Value', 'Kurtosis', 'Label'])

📊Figure 3 - Feature Separation (RMS)

🏆

4.1 Evaluation of Results

Obtained attribute data Random Forest It was trained using the algorithm.

%100

Accuracy

80/20

Training/Test Ratio

📈 Feature Importance: Your most important feature Maximum Value and RMS It was seen as a result of the analysis.

🔍 Confusion Matrix Results

	Prediction: Solid	Prediction: Defective
Fact: Solid	✓ Correct	0
Fact: Defective	0	✓ Correct

Model intact and defective parts without error has separated.

💻 Model Training Code

# Source - StackOverflow (CC BY-SA 4.0)
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import accuracy_score
from sklearn.model_selection import train_test_split

# Split Data
X = df_features[['RMS', 'Max_Value', 'Kurtosis']]
y = df_features['Label']
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42)

# Model Training
model = RandomForestClassifier(n_estimators=100)
model.fit(X_train, y_train)

# Prediction
y_pred = model.predict(X_test)
acc = accuracy_score(y_test, y_pred)
print(f"Model Accuracy: {acc*100:.2f}")

📊Figure 4 - Confusion Matrix

🛠️ Technologies Used

🐍

Python

📊

NumPy

🔬

SciPy

🤖

scikit-learn

📈

matplotlib

🐼

pandas

Bearing Fault Detection - Predictive Maintenance and Machine Learning Project