Enhancing mechanical ventilator reliability through machine learning based predictive maintenance

Area of Science:

• Clinical Engineering
• Artificial Intelligence in Healthcare
• Medical Device Maintenance

Background:

• Artificial Intelligence (AI) offers transformative opportunities in clinical engineering, including predictive maintenance of medical devices.
• Mechanical ventilators are vital for respiratory support, making their reliability and safety paramount.
• Ensuring the consistent performance of medical equipment is crucial for patient well-being and preventing adverse events.

Purpose of the Study:

• To explore machine learning techniques for enhancing predictive maintenance of mechanical ventilators.
• To evaluate the performance of various machine learning algorithms in predicting ventilator failures.
• To identify the most effective AI models for ensuring the reliability of respiratory equipment.

Main Methods:

• A dataset of 1350 mechanical ventilator entries from 15 manufacturers and 30 models was utilized.
• Machine learning algorithms including Logistic Regression, Decision Trees, Random Forest, K-nearest Neighbors, Support Vector Machines, Naive Bayes, and XG Boost were developed and tested.
• Model performance was evaluated based on accuracy in predicting mechanical ventilator failures.

Main Results:

• Ensemble methods like Random Forest and XGBoost demonstrated superior performance in handling complex ventilator data.
• Decision Tree and Random Forest models achieved high accuracies of approximately 0.993.
• K-Nearest Neighbors (KNN) exhibited near-perfect accuracy in predicting mechanical ventilator failures.

Conclusions:

• AI-driven automated systems can address challenges in maintaining the quality of medical devices in healthcare.
• Implementing machine learning-based predictive maintenance significantly enhances the reliability of mechanical ventilators.
• Predictive maintenance using AI improves patient safety and healthcare workflow efficiency.