A Bayesian Failure Prediction Network Based on Text Sequence Mining and Clustering.

Area of Science:

• Aerospace Engineering
• Data Science
• Artificial Intelligence

Background:

• Aircraft maintenance generates significant unstructured text data, often overlooked in traditional failure prediction.
• Current failure prediction models primarily rely on structured data, limiting their scope.
• Unstructured failure text contains valuable information for predicting component malfunctions.

Purpose of the Study:

• To develop a novel approach for predicting aircraft failures using textual sequence data.
• To leverage natural language processing (NLP) and machine learning for analyzing unstructured maintenance logs.
• To improve the accuracy and scope of failure prediction in aviation maintenance.

Main Methods:

• Applied NLP techniques including text segmentation and stop word removal to Chinese failure text data.
• Utilized the word2vec model to generate failure occurrence sequences from text data.
• Employed clustering algorithms for fault type classification and sequence mining (e.g., PrefixSpan) for pattern discovery.
• Trained a Bayesian failure network model using the extracted failure sequences.

Main Results:

• Successfully classified typical fault types using clustering on failure text data.
• Identified significant failure occurrence patterns through sequence mining.
• The Bayesian failure network model demonstrated superior accuracy in predicting failures compared to existing methods.

Conclusions:

• Predicting aircraft failures from textual sequence data is feasible and effective.
• NLP and sequence mining techniques can unlock valuable insights from unstructured maintenance data.
• The proposed Bayesian network approach offers a promising advancement for proactive aircraft maintenance and safety.