Natural Language Processing (NLP) applications in Maintenance and Repair

Success of each maintenance operation depends on the availability of critical parts for repairs, knowledge of the staff, published manuals, previous work experiences and availability of knowledgeable consultants among other choices. Several choices such as repair vs replace or multiple repair options exist for each maintenance operation. These choices can be ranked based on time it takes to repair and the costs involved.

Maintenance and repair of the equipment generates an assortment of structured and unstructured data. There are well established approaches to capturing and using structured data that resulted in great performance improvements. Next challenge is to leverage existing unstructured data and using it along with structured data. Such an integrated approach can have dramatic positive effect on the maintenance processes, downtime reduction and operational readiness besides improved benefits for the data quality improvement and timeliness of the data to solve a particular maintenance issue.

Current technologies that permeate Engineering and Maintenance organizations have great strengths managing structured data including documents and designs. Well-designed system exist today that store Part Metadata along with the CAD file designs, the viewable design files along with specifications, operating/repair manuals/processes, design details, test and inspection details. Part effectivities and release status make sure the correct configurations are available.

Document Management within the structured domain includes view, markup, edit and print tools. For almost a decade now, several text indexing and searching software were integrated into these systems enabling Engineers or Technicians to experience better productivity via enhanced search techniques.

Combining the structured data with unstructured data enables a firm to generate multiple repair options and select optimal option.

Importance of Emerging technologies

In recent years, Natural Language Processing (NLP) experienced rapid growth. NLP as a process has now well-defined demarcations or phases of extraction along with several effective algorithms at each phase. A well-defined consensus emerged on how to measure effectiveness of each step. Measures such as precision, recall and F-scores are widely accepted as measurement tools.

Despite the technology being “bleeding” edge, NLP techniques are finding use in the Medical field. Success of these techniques prove robustness of the NLP to extract and create knowledge based systems that can effectively solve real world problems,

NLP Solution criteria and components

NLP solution should meet multiple criteria such as modularity, extensibility, parallelizability that works well with large-scaled systems that allows work with multiple formatted data streams. Components or stage of NLP include,

  • Sentence Boundary Detector
  • Tokenizer
  • Normalizer
  • Part-of-speech tagger (POS tagger)
  • Shallow Parser
  • Named Entity Recognition (NER) annotators including negation and status annotation


  • Integration between Structured and Unstructured data gives multiple advantages including ability to select optimal maintenance process that minimizes costs and improve operational readiness.
  • Provides vital information for repair analytics that can minimize the inventories and at the same time improve operational readiness. Gives clear decisions on replacement vs repair decisions. Allows better scoring of available options.
  • Can help sorting out some of the data quality issues and help remove “bad” records from the databases.
  • Lessons learned from each maintenance can be easily shared thus allowing more effective maintenance or repair process.
  • Knowledge dissemination becomes easier. External consultation can be minimized. Knowledge preservation helps reduce equipment downtime and costs.

Product data – Why bother?

A few years back, I was sitting opposite a former executive of a manufacturing firm that supplies advanced components to aircraft industry, in lovely Santa Monica, CA. He told me of his former employer, how they were able to land an order to manufacture a complex component after several months of marketing effort. Input material for this component is an expensive alloy that required special machining techniques.

In discussions with their customer, the company finalized their designs after several iterations. At the time of manufacturing, they made a mistake! They sent a previous revision of drawing to the shop floor. Based on this, shop floor engineers machined the expensive material incorrectly.

They resumed production after fixing this mistake. Meanwhile, their schedules slipped and reordered material wiped out any profit from this project.

This is, what I call, a costly avoidable mistake. However, mistakes do happen. But, with proper controls and verifications such mistakes are very rare.

Product data management/lifecycle management systems and processes make these mistakes virtually non-existent.