The client is a leading industrial designer and manufacturer of precision machine parts offering end-to-end designing and manufacturing solutions. The company primarily engages with aviation industry part suppliers across the USA for manufacturing high-quality parts.
The manufacturing company has in-house design research and development teams creating design and drawings to enable manufacturing at its production facility. Given the aviation industry standards and heavy monetary investment for precision manufacturing, they followed stringent quality checks.
However, as the entire process was manual, it was time-consuming and resource-intensive. Thus, they looked for automation solution for QC of pdf files of drawings to reduce the time spent and costs incurred due to reworks in manufacturing.
- Lack of proper test strategy and test coverage to check drawings quality
- Wide range and uncertainty in the scope of parameters to be audited
- Difficulties in reading heavily detailed manufacturing instructions and foot notes in drawings
The company approached HitechDigital for capture and standardization of Reddit posts and discussions and classification according to race, ethnicity, age, etc.
The team at HitechDigital studied the processes to understand the scope of work, technology to be used, and the workflow to be designed.
Following project requirements increased the challenges of the process:
- Extract unstructured data in form of paragraph or text, which had no repetitive pattern.
- Identify, flag and skip comments removed by moderators – not to be collected.
- Expand sub replies and collapsed comments according to the hierarchy to collect/scrape data.
- Skip conversations that broke the continuity of discussion thread and led to another page altogether.
- Omit extraneous data such as time stamps, points, etc., which increased project complexity.
HitechDigital data analysts designed an automated quality check process for verifying information warehoused in the manufacturing shop drawings using Robotic Process Automation(RPA).
The team designed bots using UI Path and R for pixel by pixel validation of following details in the drawings and models:
- Information in the title block such as scale, drawing methods used, spell checks for the client details, etc.
- All five views in the drawings, the scale used, units of dimension and consistency
- Manufacturing details mentioned in foot notes
- Dimensioning and tolerance accuracy
- Metadata information in the case of part files in form of 3D models
Defining the scope of automation:
- The client shared the PDF files of detailed shop drawings as well as part models with the project team
- Our automation experts analyzed different approaches and technologies like Adobe image to text, Google OCR tesseract, and PDF tables in Python. However, none were competent to read the drawing details accurately
Programming for bots:
- The development team designed a system competent to read details like drawings details for geometrical accuracy, tolerance accuracy up to three decimals, spell check in title blocks etc.
- The team zeroed down on UI path, a RPA tool, to replicate the human steps on a computer for QC process by deploying a bot to identify the information accuracy, compare it against the standard values and flag discrepancies.
- The bots were trained to validate pixels on drawings to check:
- Information in the title block mentioned at the bottom of the drawing document, consistency of units such as mm or inches across the drawing, spell checks against the standard dictionary.
- Dimensions and tolerance values against the standard rule book. For example it could differentiate tolerance of 0.01 from 0.010.
- Each view in the drawing, RHS, LHS, top, bottom, and front views by comparing zone wise dimensions. Missing details in the complete isometric drawings could be highlighted.
- For part models, the bots were imparted intelligence to look for missing parameters in the metadata information either just before final packaging or at certain predefined standard intervals.
- If any discrepancy of spelling, incorrect data, or missing detail was found, the bot listed down the errors in the output excel file.
- A senior engineer performed quality checks of random sampling by verification of drawings manually and comparing it with the bots results.
- The team fixed NLP algorithms as and when errors were encountered to raise the quality bars.
Final bots were dispatched to the client for integrating with their systems.
- Software and Technology Used: UI Path, R, and Excel