Services
SERVICES
SOLUTIONS
TECHNOLOGIES
Industries
Insights
TRENDING TOPICS
INDUSTRY-RELATED TOPICS
OUR EXPERTS
AI-driven quality control enablement
The customer was looking to enhance their test equipment offering aimed at automating and improving quality control at manufacturing facilities. Itransition’s team validated the viability of AI for automated measurement of caffeine pouches and detecting deviations from predefined standards, and delivered a scalable, production-grade AI solution.
99.1%
defect detection accuracy
1.5 seconds
quality control execution time per product
60%
reduction in equipment run-in time
Engagement model
About the customer
Our customer is a large manufacturing enterprise with operations across North America. The company specializes in test and measurement solutions serving the food and beverage industry.
Domain
Manufacturing
Company type
Private
Geography
North America
The challenge
The customer’s offering included the equipment for quality control of caffeine pouches. For this type of product, precise geometry and proper sealing of all seams are essential. Any deviation from strict production requirements can cause the pouch to disintegrate when used.
The customer’s equipment already employed automation for defect validation, but it relied on traditional statistical and mathematical algorithms. With an expanding range of pouch types and brands, this approach introduced errors and significantly limited scalability, as each new algorithm variation required manual customization.
The customer therefore needed a solution that could scale efficiently without time-consuming configuration for each variation. The solution needed to be deployed on both the new testing equipment as well as the legacy one with limited computational resources.
The solution
At a glance
Itransition ran a discovery phase, built a proof of concept to assess AI feasibility for product dimension measurement and defect detection, and delivered a scalable solution ready for production deployment.
Discovery
Itransition conducted a discovery session and suggested implementing a full-cycle AI-based solution capable of automatically measuring packaging dimensions – such as length, width, edge geometry, and seal placement – of caffeine pouches, as well as detecting deviations from predefined standards. Unlike traditional algorithmic approaches, an AI-based solution offers stronger generalization capabilities and can be effectively applied across multiple pouch types.
Proof-of-concept creation
To demonstrate AI feasibility for caffeine pouch measurement and defect detection using the client’s equipment, Itransition started with a proof-of-concept solution. Since the client provided us with an extensive set of real production images for algorithm development and validation, we were able to proceed directly with model development. During the PoC phase, we used a dataset of 1.5k images to train a segmentation neural network, which we subsequently integrated into the customer’s equipment.
Upon receiving an image, the model segmented relevant parts of the packaging, enabling the system to calculate necessary dimensions, such as length, width, and seal placement. As an output, the system returned a JSON file and a marked-up image containing measurements along with defect validation results and defect location data, where applicable.
The solution was launched in shadow mode, during which the legacy and new software ran in parallel and made decisions independently. All cases where the results differed were additionally analyzed by experts. Based on analysis outcomes, our team could further refine the system. The main objective of this phase was to achieve the quality required for production deployment, followed by performance optimization as the next step.
Performance optimization
Although the PoC confirmed the solution’s quality, its performance was constrained by the client’s legacy equipment with limited computational capacity. It operated on a CPU-only setup and lacked both internet connectivity and GPU support, which limited the deployment of computer vision solutions.
To overcome these constraints and achieve the required processing speed, our team implemented a series of optimizations, which resulted in a 5x increase in the solution’s processing speed:
- Image compression to reduce the size of input files and accelerate their processing
- Model quantization to reduce computational load and improve the solution’s responsiveness
- Output configuration mode, allowing the client to analyze the different set of parameters, depending on their operational needs
In the end, despite hardware limitations, the PoC successfully demonstrated the viability for AI-based inspection.
Full-scale solution delivery
Following the PoC, the client significantly expanded their dataset to include a broader range of packaging designs, introducing more brands, complex geometries, and non-standard cases. After achieving the required quality and performance on one machine operating in shadow mode, this machine was transitioned to production use with human oversight. After one month of operation, the customer’s quality control team approved the AI solution for production deployment. After approval, the system was rolled out across all of the customer’s machines in four stages. Each stage included a one-month period of supervised operation, with advancement to the next stage occurring only after successful completion of the prior stage.
The new solution operated the following way:
- The conveyor belt camera captures an image
- The neural network model segments the image into four predefined regions
- The system calculates necessary measurements of each segment (dimensions and coordinates)
- Based on measurements, the model determines the characteristics of a reference pouch and evaluates each sample against this reference
- The solution returns a JSON file and a marked-up image to end-users, informing them whether defects are present and highlighting the exact location of any detected defects
The trained model could handle both standard and non-standard packaging designs, which made it suitable for full-scale deployment.
The outcome
We delivered a scalable, production-grade AI solution, developing it from a feasibility-focused proof-of-concept and overcoming hardware constraints, helping the customer achieve the following improvements:
01
Standard and non-standard pouch geometries support by replacing variant-specific statistical algorithms with a generalizable AI model
02
Streamlined quality control, with 99.1% defect detection accuracy and 1.5-second execution time per pouch
03
Up to 85% decrease in equipment configuration effort and associated costs when adding new pouch types and brands
04
60% reduction in equipment run-in time, accelerating the transition from initial setup to stable operations without missed defects
Related cases
Service
Artificial intelligence services & solutions
Itransition offers full-cycle artificial intelligence services to help companies build and scale powerful AI solutions tailored to their business needs.
Insights
AI in manufacturing: top use cases, trends & implementation guidelines
Explore key applications and adoption benefits of AI in manufacturing, current technology trends, and best practices for overcoming implementation roadblocks.
Case study
AI solution for clinical encounter documentation automation
Itransition developed an AI-powered add-on for Microsoft Cloud for Healthcare to streamline patient encounters and provide better quality documentation.
Case study
AI consulting for plywood defect detection & grading
We provided consulting services to a wood products manufacturer, helping them implement AI for defect detection to improve efficiency and client satisfaction.
Case study
AI solution for medical packaging defect detection
We built an AI defect detection solution that ensures 100% inspection of medical containers’ coating and helps the company maintain high production throughput.
Case study
AI answer engine for doctors and patients
Learn how we delivered a PoC of an OpenAI-based web application that answers medical questions interacting with the Davinci AI model.
Case study
Digital transformation for a food manufacturer & distributor
See how Itransition implemented Dynamics 365 Finance & SCM for a food manufacturer and distributor to streamline processes and boost operational efficiency.
Case study
Dental-orthopedic order management solution
Learn how Itransition substituted multiple legacy solutions with a custom centralized order management system to improve workflow efficiency and alignment.