Fuzzy Logic for Fault Diagnosis in Industry

Inspiration

Industrial machines fail unexpectedly, causing downtime, safety issues, and financial loss. I wanted to build something that can detect faults early, even when the data is noisy or unclear. Fuzzy Logic combined with Machine Learning felt like the perfect way to create a system that thinks more like a human while maintaining the accuracy of modern models.

What it does

Our project predicts whether industrial equipment is in a normal, warning, or faulty condition.
The Machine Learning model detects hidden patterns in sensor data, and Fuzzy Logic refines the output with human-like reasoning to make the final decision more reliable.

How we built it

• Collected and cleaned industrial sensor data
  
• Performed feature engineering and normalization
  
• Trained an ML classifier (Random Forest / SVM / etc.)
  
• Designed Fuzzy Logic membership functions and rules
  
• Combined ML predictions with Fuzzy inference
  
• Visualized results and tested performance

Challenges we ran into

• Handling noisy and incomplete sensor data
  
• Deciding the best fuzzy membership functions
  
• Integrating ML predictions into a fuzzy rule-based system
  
• Ensuring stability and accuracy across multiple fault types

Accomplishments that we're proud of

• Successfully integrating Fuzzy Logic with ML
  
• Achieving higher accuracy than a standalone ML model
  
• Building an interpretable system that industries can trust
  
• Creating a working prototype under hackathon pressure
  

What we learned

• How hybrid AI systems outperform single models
  
• Better understanding of industrial sensor behavior
  
• Practical experience with fuzzy systems and rule design
  
• End-to-end ML pipeline from preprocessing to evaluation
  
• Teamwork, fast debugging, and building under deadlines

Built With

• fuzzy-logic-(skfuzzy)
• jupyter-notebook
• matplotlib
• numpy
• pandas
• python
• scikit-learn
• streamlit