Inspection Robotics – The Next Frontier for Robotics

Published on : Saturday 05-09-2020

The coming period shall witness an explosion of new solutions and its providers in the Inspection Robotics space, says Sandeep Dawkhar.

As we progress to the next phase of Robotics & Automation in the Indian industry, the next frontier to watch out for would be Inspection Robotics. We do have several solution providers in India working in this space already and specialised solutions being provided but what we will see in the coming period is an explosion of new solutions and its providers in this space. So, let us try to understand the major technologies used in Inspection Robotics, the methodology of robotic inspection usage across major industry sectors, the process flow and the major advantages of inspection robotics.

As I mentioned earlier, inspection robotics is not something which is recent. Till now we have seen niche working in this space with specialised solutions. What’s changed now is the extensive ability to use Artificial Intelligence or AI and more specifically Deep Learning techniques to train a neural network by recording a large volume of examples of what a regular component or item being inspected would look like and what a damaged one would look like. The robotic inspection system would then use the training or learning during actual production inspection. Given the way robotic automation capabilities are increasing especially with the advent and technology progress in collaborative robots, cobots with AI is the right combination for manufacturing companies aspiring to automate their inspection operations.

Types of robotic inspection technologies

Robotic inspection technologies can be classified into two clear categories:

1. Remote Inspection Robots: These are robots which can be deployed in the field for inspection activities and also in remote locations like oil rigs, power generation facilities, wind turbines, solar farms etc. Such type of robots can be wheeled or aerial like drones with a capability to travel to the remote facilities and equipped with requisite scanners, etc., to enable them to perform for the inspection activity. These robots also tend to be specialised compared to fixed inspection robots since they are usually developed for a very specific and niche inspection activity.

2. Fixed Inspection Robots: These are the robots which are currently deployed in industry at manufacturing stations and can be articulated robots or cobots equipped or fitted with the necessary inspection devices like cameras, laser equipment, scanners, etc., to enable them to perform the inspection activity. These robots are more generalised compared to remote inspection robots and do have the capability of multiple deployments with minimal changes primarily in the AI algorithm depending on the features or attributes being inspected.

Methodology of Robotic Inspection across major sectors
The Manufacturing sector: The manufacturing sector uses a combination of both remote and fixed inspection robots for their activities. In order to keep the manufacturing set-up in continual running mode (especially for process industries which cannot afford to halt certain manufacturing processes), deployable robots play a key role by travelling to the machinery and checking for equipment damages like cracks, failures, loss of performance, etc., and reporting back for remedial action. Articulated robots and cobots with inspection devices are usually used for production quality inspection and assurance activities for identifying defects like cracks and other surface defects, for optical character recognition (OCR) or for
dimensional inspection. These robots typically use a high content of AI or deep learning for performing the inspection application.

The Energy sector: The energy sector which is responsible for energy

generation/creation/exploration and then subsequent distribution typically is a major user of remote inspection robots which can travel to the remote locations where it’s not only difficult for a manual inspection activity to be performed but also dangerous given the context of inspection. Robotic inspection in this sector is also crucial for equipment reliability and up-time. It is now very common to use inspection robots for checking underwater structures in oil rigs, visual inspection of pipes for corrosion, leaks and other failures. Drones were already being used for detecting damages to panels in solar farms where it is impossible for a human to inspect and detect when you have the farms stretched across inhabitable land. What has upgraded in terms of technology is the capability to use AI algorithms for recognising the type of defects and damages and planning a more focused action instead of a generalised action.

The Transportation sector: The transportation sector typically consisting of transport solutions like ships, aircrafts and land mobility vehicles now use a very high level of inspection robotics for their applications. Like the energy sector, this sector too is a prominent user of remote inspection robots. Robots equipped with necessary inspection equipment can inspect ship hulls, aircraft bodies. Inspection activities, especially of aircraft, can be critical and complex and inspection robotics help to address this requirement by providing consistent and definite results which humans cannot provide.

Process flow for Robotic Inspection with AI
The following would be the typical process flow for implementing a robotic inspection solution with AI:
1. System preparation for capturing the images of the part being inspected
2. Image acquisition using the right combination of cameras and lights
3. Image enhancement or filtering or removing noise from the system
4. Features extraction or processing of the image
5. Defect detection and classification
6. Analysis using the standard machine learning algorithm, and
7. Final inspection results for customer.

For developing the Deep Learning algorithm for predictive inspection, one will need to however capture thousands of images and go through the following process:

a. Gather training data
-Capture images
-Perform image annotation.
b. Train the model
-Pre-processing using the GPU
-Model training using GPU.
c. Image prediction
-Deploy the model at Customer and capture new images
-Image prediction using the deployed model and inspection results for Customer.

When to use what?
There is always a thought on what to use when, i.e., a traditional machine vision inspection solution vs a deep learning-based inspection solution. The following explanation would help clear the thought:

Traditional machine vision: Usually used for barcode reading and identification, presence and absence of features in the component being inspected or for simple gauging and measurement.

Deep learning-based inspection solution: Usually used for cosmetic defects, classifying different materials based on texture, complex OCR and associated inspection projects. There is sometimes a thin line in the applicability of different techniques for some projects which is best judged by the AI engineer based on the complexity of the project. One should always remember that implementing a deep learning-based inspection solution
will demand a huge amount of data in terms of images to create the deep learning Convolutional Neural Network (CNN). The challenge is thus to convince Customers that they have to support in providing the requisite number of components with stated defects for developing the deep learning model for a reliable solution with predictive Inspection capability.

Major Benefits of introducing Robotic Inspection:
1. Accessibility: Inspection robots can access locations which are not easily accessible by humans. Robots can be designed to be nimble with a smaller footprint to access such areas.
2. Safety: Inspection robots can be deployed in hazardous locations where humans cannot operate. Inspection activities in hazardous zones can be expensive given the investment in safety kits and equipment required by a human to access such zones. Inspection robotics help eliminate this expense and also help prevent loss of human life.
3. Efficiency and Reliability: Repetitive inspection can be monotonous for humans and can bring the element of error if a human is working in a particular environment for a significant amount of time. Robotic inspection guarantees the efficiency of the inspection process and is far more reliable than human inspection.

4. Subjectivity & Perception: Humans are prone to be subjective specially when they are inspecting to a standard and there tends to be a variance in understanding a standard of what one human perceives as OK compared to the other human inspector. Robotic inspection with Deep learning help removes the subjectivity from the inspection process and delivers consistent and repetitive results over time.

5. Cost: The acquisition cost of an inspection robot is high especially when you have AI enabled inspection solution but the RoI is quick when one compares in the context of not only elimination of human labour but also savings from no/less investment in safety kits and equipment which is necessary for humans to operate in Industrial zones.

Inspection robotics applications with AI across industries:
1. Automotive industry: For defects identification like cracks, dents, scratches across external painted surfaces and also machined parts incl. identifying texture differences for resin and fabric, and for OCR inspection.
2. Electronics industry: For defects identification like cracks, scratches, burrs in PCBs, electronic parts and panels.
3. Construction industry: For defects identification like cracks, scratches, dirt, change in surface patterns for construction material like wood, metal, tiles and raw material like rubber, glass, etc.
4. Food industry: For defect identification like presence of foreign object/contamination and leakages in packaging.
5. Medicines manufacturing: For defects identification like presence of foreign object/contamination, packaging cracks, misprints, etc.

Sandeep Dawkhar is currently working as the COO of Griffyn Robotech Pvt Ltd. He has completed his Bachelors in Engineering in Production & Masters in Finance from Mumbai University. He is a member of PMI International and is PMP certified. Griffyn Robotech is a cutting-edge AI Innovation Company that works in Inspection Automation, Robotics, Defense Systems, Industry 4.0 and Self-Aware Systems. Griffyn Robotech has developed patented systems for Robotic Inspection using AI like DEEPSIGHT® and OPTIVITY®. Sandeep has more than 27 years of Industrial experience across functions; Manufacturing Engineering, Vendor development, Program Management, Business Development, Marketing & Sales, and has worked in companies like Mahindra & Mahindra, Bajaj Auto Ltd, ABB Robotics, Mahindra Defence Systems and TAL Manufacturing Solutions. Sandeep comes with experience of project execution and business development in varied industries; automotive, aerospace, defence, railways, metro and construction equipment.