Improving Quality with Intelligent Product Inspection


The client is a global medical technology company that manufactures and sells medical devices to health care institutions, science researchers, clinical laboratories, the pharmaceutical industry, and the general public.

The company partners with organizations and governments around the world like Australia, South Korea, and UAE to name a few, and have associates across 50 countries. The company is the chosen partner to distribute large quantities of syringes and needles for Covid -19 vaccination in many countries.


The client’s manual inspection of printing labels on syringes, syringe wrappers, and in-line clearance led to significant errors in the printing, higher costs from lower workforce efficiency. One mistake in the printing of a sample meant discarding of the entire LOT/batch of syringes. This was a considerable problem especially since the client was the chosen government partner for the distribution of large quantities of syringes for covid-19 vaccination.

Product recalls were costing the loss of hundreds of thousands of dollars for clients. The clients also experienced the potential loss of faith of customers and distributors in the brand.

Client Challenges 

Manual Inspection is labour intensive, tedious and prone to errors

Prior to VisionAI implementation, the process of visually inspecting pharmaceutical labels was performed by human operators.

  • Each label has to be carefully scrutinized to ensure that the print is clear, eligible and the printed information such as MFG date, expiry date, LOT number, directions for use and pricing validation are accurate.
  • Occasionally, the printing process generates defects such as broken characters, smudges and ink splashes thereby creating errors in the labels.
  • Gross printing defects that occurred across several samples were detected manually, but small critical errors on other labels were missed out.
  • Resolving these issues means checking each and every syringe and label at low speeds, which would add up time consumption and labour.

How VisionAI Made the Difference

Vision AI was integrated into the existing production line system to read characters, and icons printed on syringes, Shell Pack/Case Pack, and Unit Pack.

  • The AI algorithms then further enhanced the resolution of each scanned image, reduced noises, and compared each label against standard accurate reference templates.
  • Defects in printing were identified in real-time and displayed to the operator for removal. Graphic templates that consistently rendered poor-quality printing were identified to further enhance quality. Since samples were not available during the development, we used representative samples and fine-tuned the system when actual samples were available.
  • The Computer Vision implementation automatically detected all the anomalies along with image evidence.
  • An anomaly report was made available, which can be downloaded in various formats. Thus, by automating line clearance , multiple levels of tedious manual quality checks were made faster and more efficiently.

Tech Edge – Vision AI Challenges and Wins

Typical neural networks with multiple pipelines  for object detection, text and texture analysis require training with substantial amounts of data and fall short inaccuracy of detection particularly in case of variants.

A platform like Cogniphi AIVI supported by multiple Vision AI engines need extremely low data for training. Moreover, there is a continuous Improvement in Accuracy with reinforced self-learning.

For the particular project, AIVI’s detection based on multiple engines AIVI Core, AIVI text, AIVI Pattern, derive co-relations and valuable insights from unstructured data ensure high accuracy compared to general ML approaches.

  • AIVI Core– A specialized engine trained, taught, and modelled to detect objects that can be moving with occlusions or inclusions.
  • AIVI Text– This engine focuses on extracting data from prints and categorizes data to train the model to interpret and analyze this data.
  • AIVI Pattern– A second-order functionality that can take additional contexts and detect occurrences of patterns. It uses a series of de-noising and signal enhancement algorithms to detect patterns from features)


  • Large sets of SKUS and its variants
  • Limited samples for modeling and training of the system
  • Type of wrapper material – Plastic coverage caused a lot of reflection problems
  • Folding on the printing areas making it difficult to read and process the data
  • Printing not consistent and compliance failures
  • Compliance checks and GOVT. norm adherence risk

VisionAI Wins

  • Automated Line Clearance Inspection with very minimal Manual Intervention.
  • Four levels of quality check that are done manually can be done with ease in a single click.
  • Devised Edge solution that runs on CPU without need for any accelerators
  • Created custom GAN to improve training data and also to predict QA issue vs lighting/orientation issues.
  • Quality Check of SKUs in production line can be done in few minutes which resulted in improved productivity by around 35%.
  • Higher Quality-End to End checking of wrappers, Shelf pack, and Case pack with the reference data (Graphics file).
  • The system detects even the minute variations against Graphics compared to the manual mode of inspection resulting in reduced defect leakage to production (reduction in compliance defects by 20%).
  • Covers 134 SKU – Unit pack and Shell Pack and Case Packs. Verification happens in line and in real-time.
  • Covers even customized samples.
  • Onboarding and Deployment of SKU in quick time.
  • Visualization through dashboard and reports.
  • Anomaly reports along with evidence for future reference.

Vision AI – Key takeaways

Visual inspection with AI operates across a wide range of industries and use cases, potentially saving manufacturers’ millions of dollars at each facility.

  • Accuracy: Computer Vision achieves a significant amount of accuracy in packaging within the acceptable criteria set by the owner’s standards.
  • Reduced Costs: With automation in computer vision, the company’s working personnel are no longer needed for the physical inspection of packaging and freeing up more working hands for important tasks. The CV also reduces human error factors, which avoid a lot of wastage.
  • Reduced Downtime: With automation, quality control became much easier and smoother, allowing for 24/7 inspection time and no rest breaks needed.
  • Repeatability: Monotonous tasks are literally what computer vision solutions were developed for. Repetitive work is done at a steady pace with no time wasted on thinking as the algorithms are already set. With computer vision, there will be no requirement for training or retraining personnel either.

Developing the Smart Factory

Client Background 

The client is a leading  automobile parts manufacturer. The company has an almost eight-decade-long legacy in the industry and apart from tyres, the company deals in in various other rubber-based products such as treads, tubes, conveyor belts, and toys. The client approached Cogniphi for solutions to address challenges related to production loss, inventory management, and material wastage that were  impacting their throughput and efficiency across the factory floor.

Our team of AI experts travelled to the company’s plant to understand the exact nature of these issues that they were facing and to design a framework of AIVI-based solutions. After a careful assessment of the underlying issues and requirements, we designed and implemented a vision-intelligence-based solution framework to help the client overcome the challenges they were facing on the factory floor.

Challenges and Solutions

Production Shortfall

It is crucial that daily production output targets are met in a factory setup. The client was facing recurring shortfalls and daily production targets were not being met. The first step was to identify the cause of the problem; Cogniphi’s team used the factory’s existing CCTV camera network and enabled it with AI-powered vision intelligence (AIVI). What followed was a real-time analysis of the video output and identification of key problem areas. Unscheduled machine stoppage was eventually identified to be the chief cause of daily target shortfall.

  • Machine downtime: This can happen due to a host of reasons ranging from improper insertion/alignment of raw material to human error. In most instances, floor supervisors and factory personnel are alerted to machine stoppages several minutes after the problem has occurred and this leads to loss of precious production time.
  • However, with Cogniphi’s Vision Intelligence blended with the client’s internal system, machine stoppages could be detected preemptively and real-time alerts would ensure that frequency and durations of machine stoppages reduced by nearly 35%.


Process Optimization 

Manufacturing processes are often drawn up in boardroom meetings far from factory floors and this can lead to disconnect between on-paper processes and the actual reality in a factory setup. Hence, processes that might seem logical on paper might not be viable in reality.

  • Component changeover: This is a common cyclical process in any production facility and it should ideally take no more than 30 seconds for the process to be completed. However, at the client’s main factory unit, a disruption in changeover processes was leading to several minutes of delays.
  • With machine status and predictive data made available to the supervisors and plant managers, the component changeover process time was optimized.
  • Material Misalignment: With the help of data from highly sensitive sensors in machines and video feed from AI-enabled CCTV cameras, our team identified that there was a problem in the alignment of the rails due to which the input material was not fitting into the machine. Factory workers often tried to forcefully push a material into the machine not knowing that the real fault lied in the railing’s alignment. Continuous monitoring for material misalignment and immediate flagging of the anomaly now enables the factory managers to fix the issue on a priority basis.

Inventory Mismanagement

A factory floor is labyrinth space with multiple production zones all replete with machines and equipment. Tracking raw material movement from storage to usage can thus be a challenging task but essential task.

  • Smart Tagging: For our client, misplaced raw material or incorrect storing of rubber rolls and other such raw material components was a challenge. For this, our team devised a Smart Tagging solution whereby raw materials were tagged with smart chips that could help with object tracking and object detection in real-time.  These tagged components when loaded onto a machine that would lead to a flashing of green (for correct) or red (for error) light on large screens hanging above conveyor belts. This way, factory workers — who are often unskilled seasonal labourers — could understand if they have loaded the right material onto a machine and give them a chance to rectify their own mistake without having to involve a supervisor.



Value delivered by Cogniphi

  • 16% increase in the company’s productivity.
  • 26 % decrease in the losses due to unmet production targets.
  • 35% reduction in unscheduled stoppages in the production of tyres.
  • 80% increase in material movement and handling compliance
  • Enhanced quality check processes due to advanced texture detection and fewer instances of raw material wastage and faulty finished products due to predictive intelligence systems


  • Integration of 1100 Active Sensors & 500 legacy cameras for detecting/predicting instances that can cause productivity loss, wastage, inappropriate handling of inventory, missed inventory, NCM movement, and operator availability.
  • Vision data, inferences, prediction, productivity and live view were presented to the plant team supervisor in a simple and effective dashboard.
  • 500 cameras processing roughly 30 GB of data per second used to create a functioning setup.
  • Server setup within the factory premises, leading to all data being processed locally and company information staying protected through internal firewalls.
  • >99% accuracy achieved for visual detection models.
  • Real-time and non-intrusive solutions designed for all problems encountered during the manufacturing process.
  • Custom data generation through Vision AI Engines reduced the need for training data.
  • Company’s existing processes were built upon and fortified, instead of complete overhaul.