材料大气环境腐蚀试验方法与评价技术进展

徐迪, 杨小佳, 李清, 程学群, 李晓刚

Review on Corrosion Test Methods and Evaluation Techniques for Materials in Atmospheric Environment

XU Di, YANG Xiaojia, LI Qing, CHENG Xuequn, LI Xiaogang

表1 应用于腐蚀研究的数据挖掘方法介绍

Table 1 Introduction to data mining methods for corrosion research

Methods Functions Advantages Disadvantages Bayesian Network Correlation analysis Exploring the causal relationship between the factors influencing corrosion Need a certain amount of data to ensure the credibility of the model Grey Correlation Analysis Correlation analysis Finding the key factors affecting the corrosion mechanism of materials Does not reflect the general law of material corrosion Random Forest Correlation analysis Quantify the size of the effect of each corrosion factor on corrosion Analysis results are limited and do not reflect the general law of material corrosion Grey Prediction Predict Minimum sample size requirement Unable to respond to the effect of other factors on corrosion Multiple Linear Regression Predict Visually describe the effect of each factor on corrosion rate Limits the use of material corrosion data Artificial Neural Network (ANN) Predict Better prediction accuracy than multiple linear regression method Overfitting can occur with small sample sizes Support Vector Machine (SVM), Support Vector Regression (SVR) Predict Prediction accuracy higher than a rtificial neural network There is a dependence on the sample size for model building Monte Carlo simulation at the macro scale Predict Application to service safety assessment of pipeline type engineering facilities Large demand sample size, unable to adjust inputs and outputs for flexible forecasting Markov Chain Predict Suitable for mining continuous, time-series data Weak handling of discrete data Random Forest Predict Corrosion data suitable for high-speed variation characteristics High dependence on sample size for model building