Globally, Machine Vision has been a popular tool for automated visual inspection in Manufacturing. In India, more recently there has been a tremendous increase in adoption owing to increased system integration expertise and understanding of the technology. With the recent advancement in Artificial Intelligence and increased computation power coupled with advances in algorithm development, Machine Vision has taken on a new avatar in the form of Deep Learning-based inspection systems. These systems are easy to train and teach and reduce the integration complexity when it comes to “teaching” the machines (aka Machine Learning) on what to look for.
However, it’s important to understanding functionally, how this technology can be applied in manufacturing. There are many different application groups. It’s important to understand the kind of application your requirement falls under so as to decide which kind of system design and technology you need to invest in. Often time there will be a need to have one (or even more) functional requirements based on your application need. . Listed below are the primary categories.
1. Object Detection
Here the objective is to locate or detect whether an object of interest is present or absent in a given image. An example is whether the egg tray or ice tray is present in a refrigerator. The vision system simply identifies the part by virtue of a “golden image” or a “pattern” that has been pre-trained which it uses to compare it with the real time images from the camera.
Measurement applications as the name suggests involves an object dimensions to be accurately determined. This is done by locating certain points on an image and measuring geometrical dimensions (distance, radius, diameter, depth etc) from this image. Examples of such applications are measuring the inner diameter of an engine cylinder bore. Another example is measuring the liquid fill level within a bottle. Measurement can be done using either 2d or 3d cameras.
3. Flaw Detection
Flaw detection applications detect abnormalities such as surface defects, dents and scratches on a product surface. Flaw detection applications need to be carefully objectified in order to ensure “acceptable” flaws can be distinguished between unacceptable flaws. The use of Artificial based machine vision is ideally suited for these applications as the system is taught based on examples rather than “rules”.
4. Print Defect Identification
Identification of printing anomalies like incorrect color shades or where parts of the print is missing or blemished is the objective of print defect identification. In these applications a golden or master image is trained to the system in order to identify any deviations from this master.
Identification using Machine Vision involves identifying a part or product in order to track this part across the manufacturing or logistical process or to verify that the right part is being produced. Identification can be done either by reading characters (OCR) or barcodes.
Locating objects is a common use of machine vision for applications such as robotic guidance. Here the machine vision system’s objective is to locate the coordinates/position of an object of interest. This information can be used to pick up the object or perform any other process that is dependent on this location. This type of machine vision application, requires the child part of interest needs to be taught to the machine vision system, to identify this part during production.
Counting as the name suggests is the use of machine vision to count objects of interest (say pharma vials in a bin or piston rings in a stack)