Predictive maintenance with convolutional neural networks

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current predictive maintenance technologies face challenges in effectively utilizing telemetry data to generate accurate and interpretable maintenance predictions for large-scale monitored systems, such as heating, ventilation, and air-conditioning systems, particularly in learning from past malfunctioning and maintenance history logs.

Innovation Solution

The development of predictive maintenance convolutional neural networks that process telemetry data from sensor devices to generate maintenance predictions and extract explanatory metadata, enabling automated, efficient, and interpretable predictive maintenance by training on cross-temporal analyses of historical data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional predictive maintenance methods are used to process telemetry data, then the system can provide maintenance predictions, but the accuracy and interpretability of predictions deteriorate when dealing with large-scale monitored systems

Engineering Contradiction:
Improveprediction accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces traditional mechanical predictive maintenance methods with a neural network-based system. The neural network automatically learns patterns from telemetry data without requiring manual feature engineering or complex rule-based systems, thereby improving prediction accuracy while managing system complexity through automated learning.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent introduces an attention mechanism as an intermediary between the neural network processing and the final maintenance predictions. This attention mechanism generates heatmaps that highlight important features in the telemetry data, providing interpretability without requiring complete redesign of the prediction system.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If comprehensive telemetry data from multiple sensors is collected to improve prediction accuracy, then the prediction effectiveness improves, but the difficulty of processing and analyzing the data increases

Engineering Contradiction:
Improvemaintenance prediction reliabilityVSAvoiddata processing difficulty
Core Design Contradiction:
ReliabilityVSDifficulty of detecting and measuring

Solution Approach 1:

The patent creates a universal neural network model that can process telemetry data from multiple different sensor types simultaneously. The model is designed to handle diverse input data formats and sensor configurations through a unified architecture, reducing the difficulty of processing comprehensive multi-sensor data while maintaining high prediction reliability.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Adaptability or versatility

If historical malfunctioning and maintenance history logs are extensively utilized for training, then the model's ability to learn desired configurations improves, but the time and computational resources required for training increase

Engineering Contradiction:
Improvemodel learning capabilityVSAvoidtraining time
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent performs preliminary data processing and feature extraction during the data preparation phase, organizing historical malfunctioning and maintenance history logs into structured formats before training. This preliminary action reduces the computational burden during actual training, allowing the model to learn desired configurations effectively while minimizing training time.

Inventive Principle:
Principle #10Preliminary action

4Productivity

If automated predictive maintenance is implemented to improve efficiency, then productivity improves, but the loss of interpretable information about maintenance needs increases

Engineering Contradiction:
Improvemaintenance efficiencyVSAvoidexplanatory information
Core Design Contradiction:
ProductivityVSLoss of information

Solution Approach 1:

The patent implements a feedback mechanism where the attention-generated heatmaps provide explanatory information back to users about which features influenced the maintenance predictions. This feedback loop maintains interpretability while preserving the automation benefits, allowing users to understand the reasoning behind predictions without manual analysis.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP4118503B1Predictive maintenance with convolutional neural networks
Publication Date: 2024.05.01 HONEYWELL INTERNATIONAL INC
  • EP4118503B1 patent drawingFigure 1
  • EP4118503B1 patent drawingFigure 2~3
  • EP4118503B1 patent drawingFigure 4~5

AI summary

The application relates to a computer-implemented method comprising: receiving, from a query-initiating device configured to display a user interface, a predictive telemetry data object, the predictive telemetry data object associated with a plurality of predictive measurements, each predictive measurement of the plurality of predictive measurements associated with a sensor device of one or more sensor devices in a monitored system; generating, based on the predictive telemetry data and one or more models of the monitored system, a maintenance prediction for the associated sensor device; generating one or more maintenance need notification outputs based on the maintenance prediction; and causing the query-initiating device to display the one or more maintenance need notification outputs using the user interface.