Case study – AI Tools for Controlling Anodized Rivets

Case study - AI Tools for Controlling Anodized Rivets

AI and measurement precision, the challenges in controlling anodized rivets

Designed specifically for the quality control of anodized rivets, the sorting machine in this case study is an indexed metal-plate rotary table machine, model MCV1, equipped with dedicated stations and software.

In addition to the prerequisite of a selection rate of no less than 100 pcs/minute, the context poses 3 challenging controls.

  1. Rivet Body diameter measurement with mandatory tight production tolerances to 12 μm
  2. Shank bottom hex-recess shape control to detect pieces with double broaching
  3. Marking logo defect detection on top of the head surface

During the testing phase, the situation proved to be even more complex due to the surface coating of the rivets which is delicate and subject to be easily scratched. 

Case study - AI Tools for Controlling Anodized Rivets
12 μm tolerance control

Dimac vision systems are designed to ensure precise control of production tolerances, especially for small metal parts applications, with a focus on tolerances up to 50μm. For even tighter tolerances, our control system is tailored ad hoc, incorporating measures that become more stringent as tolerances decrease. We’ve implemented a lens aberration correction algorithm, along with dedicated backlighting and a higher resolution camera paired with top-quality telecentric lenses. This combination aims to minimize the impact of glass irregularities and standardize diffraction effects at the piece’s edge across various images.To counteract the effects of environmental dirt in the production setting, our vision system includes advanced software filters for added protection and reliability.

The entire measurement process demands substantial computing power, driven by the necessity to capture 3 images per piece (one every 120°) at a rate of 100 pieces per minute, all while utilizing various software filters. To meet this demand, we employed an industrial computer equipped with an NVIDIA® graphics card. The outcome is a machine that successfully passes both MSA1 and MSA3 tests, even within the design tolerance range of 12 μm. It’s noteworthy that the increase in camera resolution brings significant benefits to measurement quality but could affect productivity. Nevertheless, the adopted solution represents the optimal balance between the mandatory controls and productivity, ensuring the finest quality-cost combination.

Checking the shank bottom hex-recess shape

The mandatory controls include: checking for the presence of the recess at the bottom of the shank, checking for the hexagonal shape of the recess and the absence of a double broaching effect (a star-like shape resulting from a slight phase shift between the first and second passage of the cutting), and checking for the charaters of the marking logo on the top surface of the head.

The primary challenges once again involve ensuring image stability, adequately highlighting all defective situations, and implementing an interpretation algorithm capable of recognizing the defects without triggering false rejections.

To address image concerns, we chose an entocentric camera with a coaxial annular illuminator directed at the bottom edge of the shank. This illuminator provides uniform lighting only when the piece is precisely cantered (with tenths of a millimeters being crucial).

Case study - AI Tools for Controlling Anodized Rivets

Leveraging the production tolerance of 12μm on the body shank-diameter, the slots on the transport dial were carefully designed to serve as the vertical constraint for hanging and guiding the piece. A spring catch positioned above the rotary table, along with the precise seat offered by the dial-slots’ chamfer to the head of the rivet, ensures the piece remains in position even in the presence of vibrations.

The outcome is an image where the hexagonal bottom recess appears prominently in black within the light circular crown of the piece’s wrench. This was good but not enough to ensure the detection of a double broached recess, because the presence of dirt could affect the control. For this reason an algorithm was developed to rely on light gradient analysis, specifically the extent of a gray band around the hexagon’s circumference. The algorithm counts how many times the gradient intersects a user-defined second circumference diameter. This measured circumference sets the limit between acceptable dirt and actual defects, allowing for adjustable control sensitivity.

Checking the characters of the marking logo 
on the top of head-surface

In the top flat head of the rivet, a variable content marking, like the production batch, is imprinted. As the piece moves under the marker, it could happen the writing appear double or blurred, making it illegible.

Image stability is crucial for this process, as any instabilities create background noise, blurring the line between compliant and non-compliant components. Traditional image analysis methods are limited and even with the aid of OCR (Optical Character Recognition) functions it is not possible to process 100 pieces/minute.

AI comes to our aid offering a solution which includes the training of a neural network to focus solely on recognizing the defect, not the textual content of the writing. The AI ignores the meaning of the controlled text but is adept at identifying peculiarities associated with the defect in the written text.

Like previous checks, the challenge lies in finding the right balance between computing times, defect discernment, and investment in component quality.

Case study - AI Tools for Controlling Anodized Rivets
Handling parts with delicate surface coating

Anodized rivets typically don’t require special handling procedures in the feeding system, except during machine setup when the piece rotates 360° inside transport disc slots. Occasionally, the side surface of the cylindrical body could get abraded during the contact between the rivet shank and the metal disk slot, this does not affect the functionality of the piece but should be prevented for aesthetic reasons. To prevent abrasions , one common method is lifting the piece off the disc, but this is not recommended here, because of time constraints. Increasing the slot diameter is not an option because of the precise stability of the piece required for the bottom hex-recess control.The final solution is an accurate centred piece rotation within the disk-slot clearance (5 hundredths of a millimetres). Overcoming this mechanical challenge requires meticulous craftsmanship in component care and assembly.

AI and Training

Artificial Intelligence (AI) control differs from traditional vision software by not relying on preset rules. Instead, the AI algorithm calculates variables, forming a neural network that assesses image similarity for categories like compliant and non-compliant parts. This network determines the likeness percentage of a new image to specific categories.

AI goes beyond analytical image interpretation, autonomously generating search criteria with logics not directly understandable by humans. While highly versatile and powerful, its results aren’t entirely predictable beforehand, necessitating careful field validation.

The AI training process, involving a well-equipped computing station and required software licenses, also demands specific expertise.

Arriving at effective and computationally efficient solutions is crucial. The AI process demands substantial calculation time for both training and validation phases. To avoid prolonged non-productive periods of the sorting equipment, the training is not carried out on the sorting machine.

For user convenience, Dimac provides remote neural network training services. 

The machine software incorporates a straightforward procedure to assist operators in capturing sufficient images of compliant and defective samples. Dimac conducts the training as a service, refines, and validates it, and then installs the trained neural network on the machine.

Ensuring a safe 100% control of the characters logo marked on the head of the rivet poses a challenge which cannot be solved with traditional techniques. Alongside human visual inspection, AI is currently the sole viable solution. The human eye remains essential in defining and optimizing the AI workflow today and possibly in the future.

Case study - AI Tools for Controlling Anodized Rivets

Table of Contents

CONTACT US

For more info about Dimac News

Your data will be processed according to the provisions of EU Regulation 2016/679 on the protection of personal data. Find out more: Privacy Policy Dimac