Multi-Sensor Data Fusion for Remaining Useful Life Prediction of Machining Tools by IABC-BPNN in Dry Milling Operations

Min Liu, Xifan Yao, Jianming Zhang, Wocheng Chen, Xuan Jing, Kesai WangSensors 2020 · DOI

TL;DR

This paper proposes a novel multi-sensor data fusion system for predicting the Remaining Useful Life (RUL) of machining tools during dry milling operations. By combining data from vibration and force sensors, employing advanced signal processing and feature selection, and utilizing an optimized neural network (IABC-BPNN), the system achieves highly accurate and reliable RUL predictions, outperforming existing methods.

What Problem Does This Paper Address?

Machining tools are critical components in manufacturing, and their unexpected failure or premature replacement leads to significant downtime, increased costs, and reduced product quality. Accurately predicting the Remaining Useful Life (RUL) of these tools is crucial for efficient maintenance scheduling and ensuring operational reliability. Traditional methods often struggle with the complexity and non-linearity of machining processes, the abundance of sensor data, and the inherent uncertainties in tool wear mechanisms. This paper aims to address these challenges by developing a robust multi-sensor data fusion system that can accurately predict RUL in real-time.

Methods

The proposed system employs a multi-stage approach:

  • Multi-Sensor Data Collection: Vibration and force sensor data are collected during dry milling operations.
  • Signal Preprocessing (CEEMD): Complementary Ensemble Empirical Mode Decomposition (CEEMD) is used to denoise the raw sensor signals, effectively handling noise and preserving important signal characteristics.
  • Feature Extraction: A comprehensive set of features is extracted from the denoised signals across three domains:
  • Time Domain (TD): Statistical features like mean, maximum, RMS, variance, etc.
  • Frequency Domain (FD): Features derived from power spectrum density analysis.
  • Time-Frequency Domain (TFD): Features obtained through Hilbert-Huang Transform (HHT) on the intrinsic mode functions (IMFs) from CEEMD.
  • Feature Selection (LightGBM): A Light Gradient Boosting Machine (LightGBM) is employed to identify and select the most relevant features that strongly correlate with tool wear, reducing dimensionality and improving model efficiency.
  • Feature Fusion and RUL Prediction (IABC-BPNN & PCF):
    • An Improved Artificial Bee Colony (IABC) algorithm is used to optimize a Back Propagation Neural Network (BPNN). This IABC-BPNN model fuses the selected features to build a tool wear prediction model. The IABC algorithm enhances the BPNN's ability to find optimal weights and thresholds, preventing local optima and improving convergence.
  • A Polynomial Curve Fitting (PCF) method is then used to predict the online RUL based on the tool wear predicted by the IABC-BPNN model.

Datasets

The study utilizes data collected from experiments conducted on a CNC milling machine. Specifically:

  • Offline Data: Data from five milling tools (C1-C5) were used. Tools C1 and C3 were measured by a microscope for offline tool wear data, serving as training and testing sets. Tools C2 and C4 were used for predicting tool wear.
  • Online Data: Tool C5 was used for online RUL prediction.
  • Sensor Data: Vibration sensor (M69) and force sensor (Kistler 9257A) data were collected at a sampling frequency of 1 KHz.

The paper does not introduce a new benchmark dataset but rather uses experimental data generated under controlled conditions.

Key Insights and Ideas

  • Synergy of Multi-Sensor Fusion: Combining vibration and force sensor data provides a more comprehensive understanding of the machining process and tool wear than single-sensor approaches.
  • CEEMD for Robust Preprocessing: CEEMD proves effective in denoising complex signals from machining operations, preserving crucial information for subsequent analysis.
  • Hierarchical Feature Engineering: Extracting features across time, frequency, and time-frequency domains captures diverse aspects of tool wear, leading to richer feature sets.
  • LightGBM for Efficient Feature Selection: LightGBM's ability to identify high-importance features significantly reduces the feature space, leading to more efficient and accurate models.
  • IABC-BPNN for Optimized Prediction: The IABC algorithm's enhancement of BPNN effectively balances exploration and exploitation, leading to superior optimization of network parameters and improved tool wear prediction accuracy.
  • PCF for Direct RUL Estimation: Polynomial curve fitting provides a straightforward method to translate predicted tool wear into a direct RUL estimate.

Key Results

  • The proposed IABC-BPNN model demonstrated superior prediction performance compared to other methods like RBFN, BPNN, and NFIS.
  • Statistical performance metrics showed significantly lower RMSE and MAPE for the IABC-BPNN model (RMSE: 0.0024, MAPE: 0.0032) compared to its rivals.
  • The R² value for the IABC-BPNN model was exceptionally high (0.9953), indicating an excellent fit and predictive capability.
  • The experimental results showed that the predicted values were within a 5% error margin of the measured values, confirming the model's reliability.
  • The system successfully predicted tool wear and RUL for different tools (C1-C5), demonstrating its generalizability.

Limitations and Future Work

  • Single Working Condition: The current IABC-BPNN model is designed for predicting tool wear and RUL under a single working condition.
  • Future Work:
    • Multi-Working Conditions: Extend the method to predict tool wear and RUL under multi-working conditions and for different types of machining tools.
  • Automatic Parameter Optimization: Further improve the method through automatic optimization of parameters.
  • Integration with Health Monitoring: Combine RUL prediction with broader health monitoring systems for machining tools.

Summary updated

Abstract

Inefficient remaining useful life (RUL) estimation may cause unpredictable failures and unscheduled maintenance of machining tools. Multi-sensor data fusion will improve the RUL prediction reliability by fusing more sensor information related to the machining process of tools. In this paper, a multi-sensor data fusion system for online RUL prediction of machining tools is proposed. The system integrates multi-sensor signal collection, signal preprocess by a complementary ensemble empirical mode decomposition, feature extraction in time domain, frequency domain and time-frequency domain by such methods as statistical analysis, power spectrum density analysis and Hilbert-Huang transform, feature selection by a Light Gradient Boosting Machine method, feature fusion by a tool wear prediction model based on back propagation neural network optimized by improved artificial bee colony (IABC-BPNN) algorithm, and the online RUL prediction model by a polynomial curve fitting method. An example is used to verify whether if the prediction performance of the proposed system is stable and reliable, and the results show that it is superior to its rivals.
BibTeX
@inproceedings{liu2020multi,
  author = {Min Liu and Xifan Yao and Jianming Zhang and Wocheng Chen and Xuan Jing and Kesai Wang},
  title = {{Multi-Sensor Data Fusion for Remaining Useful Life Prediction of Machining Tools by IABC-BPNN in Dry Milling Operations}},
  booktitle = {Sensors},
  year = {2020},
  doi = {10.3390/s20174657},
}
Figure