Machine Learning Failure Prediction for Complex Machine Tools

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing abnormality diagnosis methods for machine tools are susceptible to disturbances due to reliance on temporary signal data, leading to unreliable failure predictions.

Innovation Solution

A complex machine tool equipped with a machine learning unit that acquires sensor information, including vibration, acceleration, and load data, to predict failures in components like bearings and ball screws by identifying frequency bands with abnormalities through amplitude spectrum analysis, enabling proactive inspection and replacement.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If temporary signal data is used for abnormality diagnosis, then the diagnosis method is simple and quick, but the reliability of failure prediction deteriorates due to susceptibility to disturbance

Engineering Contradiction:
Improvefailure prediction reliabilityVSAvoiddiagnosis system complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The system performs preliminary actions by continuously collecting and storing sensor information over time before actual failure occurs. The machine learning unit accumulates sensor data during normal operation to establish baseline patterns, enabling more reliable failure prediction when anomalies are detected.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

A machine learning unit is introduced as an intermediary between raw sensor data and failure diagnosis. This intermediary processes sensor information, identifies abnormal patterns, and generates failure predictions, thereby improving reliability while managing system complexity through intelligent data processing.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If sensor information is collected continuously for failure prediction, then prediction reliability improves, but data processing complexity and time consumption increase

Engineering Contradiction:
Improvefailure prediction reliabilityVSAvoiddata processing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system applies partial action by focusing processing efforts only when necessary. The machine learning unit continuously monitors sensor information but performs intensive analysis only when abnormal patterns are detected, rather than processing all data at full capacity continuously.

Inventive Principle:
Principle #16Partial or excessive action

Solution Approach 2:

The machine learning unit performs self-service by automatically identifying abnormal patterns and generating failure predictions without requiring constant human intervention or complex external processing systems. The system serves itself by autonomously managing the balance between continuous monitoring and selective deep analysis.

Inventive Principle:
Principle #25Self-service

3Reliability

If machine learning unit is introduced for failure prediction, then prediction reliability improves, but device complexity increases

Engineering Contradiction:
Improvefailure prediction reliabilityVSAvoidsystem structure complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The machine learning unit is designed with multi-functionality, serving as both a data processor and a prediction engine. It handles multiple sensor types (vibration, temperature, etc.), performs pattern recognition, and generates failure predictions, thereby reducing the need for separate dedicated components for each function.

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

Solution Approach 2:

The machine learning unit acts as an intermediary layer that simplifies the overall system architecture by consolidating complex data processing and analysis functions into a single intelligent component, rather than requiring multiple separate systems for data collection, processing, and decision-making.

Inventive Principle:
Principle #24Intermediary (Mediator)

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

The solution provides highly reliable failure predictions, allowing for timely inspection and maintenance, thereby enhancing the robustness of machine tool operations and preventing unexpected failures.

Implementation Method 1

detecting vibrations due to relative rotation, converting the vibrations into an electrical signal

Methodology Applied
Scientific EffectVibration: Vibration

Implementation Method 2

the sensor information is acceleration information and the amplitude spectrum is an amplitude spectrum of a power spectrum density (PSD)

Methodology Applied
Scientific EffectAcceleration: Accelerometer

Implementation Method 3

which of frequency bands of interest involves a spectrum with abnormality is identified by using amplitude spectrum information obtained by the sensors

Methodology Applied
Scientific EffectFrequency spectrum analysis:

Data Source

PatentEP3413027B1Complex machine tool having failure predication function
Publication Date: 2021.02.24 NAKAMURATOME SEIMITSU IND
  • EP3413027B1 patent drawingFigure 1~2
  • EP3413027B1 patent drawingFigure 3~4
  • EP3413027B1 patent drawingFigure 5

AI summary

A complex machine tool comprises a sensor (41), an FFT analysis unit (112), a machine learning unit (130), and a comparison unit (160). The sensor is attached to a vibration part (50, 60). The FFT analysis unit (112) performs frequency analysis on sensor information acquired by the sensor (41). The machine learning unit (130) acquires an amplitude spectrum of a frequency band of interest obtained by the frequency analysis performed by the FFT analysis unit (112), and calculates an average amplitude spectrum of the frequency band of interest based on a learned leaning model. The comparison unit (160) compares the amplitude spectrum of the frequency band of interest obtained by the FFT analysis (112) unit with the average amplitude spectrum of the frequency band of interest obtained by the machine learning unit (130).