Predictive Maintenance for Industrial Robots

This project explores predictive maintenance strategies for industrial robots used in nuclear fuel replacement tasks. It leverages data analysis, machine learning models, and decision-making frameworks to enhance fleet management and maximize operational efficiency.

Project Phases:

1. Context: Overview of predictive maintenance requirements.
2. Environment: Description of the technical environment and dependencies.
3. Knowledge Base: Core mathematical and machine learning concepts.
4. Input Data: Details on data structure and file formats.
5. Testing: Model performance evaluation process.
6. Exploration: Initial data analysis and engineering techniques.

Technical Phases:

• Feature Engineering: Time series and statistical features derived from failure data.
• Prediction Models: Various models such as Random Forest, LSTM, and Gradient Boosting Survival Models for Remaining Useful Life (RUL) predictions.
• Performance Evaluation: Metrics including accuracy, confusion matrix, ROC-AUC, and precision-recall curves.
• Remaining Work: Development of Phase II decision-making model.
• Lessons Learned: Insights into project organization, time management, and the importance of structured version control.

Detailed Phases:

1. Phase I – Remaining Useful Life (RUL) Prediction: Predicting if a robot will remain operational over the next six months based on time series failure data.
   • Evaluation Criteria: Success metrics calculated based on actual vs. predicted RUL, with rewards and penalties for prediction accuracy.
2. Phase II – Maintenance Decision-Making: Assessing whether to replace robots before a new mission, with the goal of maintaining at least 10 operational robots.
   • Evaluation Criteria: Rewards for successful missions and penalties for robot failures or unnecessary replacements.

Data Structure:

• failure_data.csv: Summarizes failure modes and associated timestamps.
• Degradation data: Time series data tracking crack length evolution in each robot.
• Testing Data: Structured datasets for model validation based on real-world test scenarios.

Implementation: The project is built using Streamlit for visualization. Key files include:

• app.py: Main application entry point.
• dashboard/: Contains dashboard components and layout.
• data/: Contains processed data and input files.

Lessons Learned:
Insights gained include key considerations for model selection, the value of setting time-focused project goals, regular version control, and structuring the repository for collaboration.

Installation

Prerequisites

• Python 3.8 or higher
• pip (Python package installer)

Clone the Repository

git clone https://github.com/mriusero/predicitve-maintenance-on-industrial-robots
cd predicitve-maintenance-on-industrial-robots

Create and Activate a Virtual Environment (Optional but Recommended)

On macOS/Linux

python3 -m venv venv
source venv/bin/activate

On Windows

python -m venv venv
venv\Scripts\activate

Install Dependencies

pip install -r requirements.txt

Running the Application

To start the Streamlit application, run the following command from the project’s root directory:

cd app 
streamlit run app/app.py

Directory Structure

.
├── testing_data
│   └── sent_to_student
│       ├── group_0
│       │   ├── Sample_submission.csv
│       │   ├── testing_data.rar
│       │   └── testing_item_(n).csv  // n = 50 files
│       ├── scenario_0
│       │   └── item_(n).csv        // n = 11 files
│       ├── scenario_1
│       │   └── item_(n).csv        // n = 11 files
│       ├── scenario_2
│       │   └── item_(n).csv        // n = 11 files
│       ├── scenario_3
│       │   └── item_(n).csv        // n = 11 files
│       ├── scenario_4
│       │   └── item_(n).csv        // n = 11 files
│       ├── scenario_5
│       │   └── item_(n).csv        // n = 11 files
│       ├── scenario_6
│       │   └── item_(n).csv        // n = 11 files
│       ├── scenario_7
│       │   └── item_(n).csv        // n = 11 files
│       ├── scenario_8
│       │   └── item_(n).csv        // n = 11 files
│       ├── scenario_9
│       │   └── item_(n).csv        // n = 11 files
│       └── testing_data_phase_2.rar
│
└── training_data
    ├── create_a_pseudo_testing_dataset.ipynb
    ├── degradation_data
    │   └── item_(n).csv            // n = 50 files
    ├── failure_data.csv
    ├── pseudo_testing_data
    │   └── item_(n).csv            // n = 50 files
    └── pseudo_testing_data_with_truth
        ├── Solution.csv
        └── item_(n).csv            // n = 50 files

18 directories, 326 files

---

2. Training Data

• Source URL: [training_data].
• Number of Samples: 50 total samples.
• failure_data.csv:
  • Summarizes times-to-failure for the 50 samples.
  • Indicates failure mode for each sample (Infant Mortality, Fatigue Crack, or Control Board Failure).
• degradation_data folder:
  • Contains crack length measurements for each sample.
  • Each CSV file in this folder corresponds to a specific sample, with item_X matching item_id in failure_data.csv.
• create_a_testing_dataset.ipynb Notebook:
  • Generates a “pseudo testing dataset” based on training data.
  • Demonstrates data generation and can be used to develop models using pseudo-test data.

3. Testing Data and Evaluation

• Test Data Link: testing_data/sent_to_student/group_0
• Sample_submission.csv:
  A template for submitting results.