Robotics - A Survey On Oil Pipe Inspection Robotics (Revising)

Posted by Rico's Nerd Cluster on December 1, 2024

Non-Destructive Testing Methods

American Society for Testing and Materials (ASTM) and American Society of Mechanical Engineers (ASME) develops standards for pipe inspections

Different liquids could cause different types of corrosions:

  • H2S (Hydrogen-Sulfide) creates small punctures, whereas most other liquids create relatively larger corrosions.

Major Players In The Market and Their Robot

MPE Supplies

  • Features: 360 deg flip, 220 deg rotate. Front light source, backlight source, video
  • $20000
  • Location: Arkansas

Dilemma of ML in Pipe Inspection - Precision and Recall

Pipe inspection is labor-intensive. People don’t trust ML that much. This lies in the ML’s lower “recall” and “precision” rates

If the inspection agency missed a spot (false negative), it could bring lawsuits where we need to explain why the manual + AI system fails to detect that. In machine learning, this is characterized as “recall”.

If the system has too many false positives, then the system is too costly to verify. In machine learning, this is characterized as precision is low. In oil and gas, precision us termed as the “prove-up” rate:

\[\begin{gather*} \begin{aligned} & p = \frac{\text{verified defects}}{\text{total detected defects}} \end{aligned} \end{gather*}\]

Explaining why relatively low precision and recall happens is not easy, as ML systems are still largely “unpredictable” during run-time. What’s worse is many customers and the court mistakenly think that ML systems should aim for perfect detection results, which is similar in the autonomous vehicle industry. In ML, we always compare the ML system with the “best human-level” performance.

Gas Meter

  • What is LEL in Gas meter? (Lower Explosive Limit)
    • Calibration: to set a baseline of “good air”. But when you calibrate it in an H2S rich environment, it’s not good
    • “Unknown gas 100” can be resolved by powercycling
    • Some gas meters have two modes: slow and fast

Inspection with 3D LiDAR Examples