Seeing is Saving: How AI-Based Vision Inspection Boosts ROI

By Andrew McGhie, Global Business Development Director of Vision Systems at KPM Analytics

Key takeaways: 

• AI-based vision inspection significantly reduces product recalls by detecting foreign materials more accurately than human inspectors or traditional methods, potentially saving millions in recall-related costs.
• Labor savings and improved efficiency come from automating tedious, high-turnover inspection tasks, with many companies achieving ROI in under a year through reduced errors and reallocated labor.
• Real-time process control powered by AI helps reduce waste and optimize production by identifying process deviations early, leading to substantial material and energy savings.

Even a few years ago, the idea of artificial intelligence having a critical role in a food production plant likely seemed far-fetched, but it is remarkable just how quickly the landscape has changed. As many food industry executives can agree, quality assurance roles at food manufacturing plants have notoriously had high turnover rates, especially product inspection roles, which, while very important in every plant, are often mundane and tedious work. The COVID-19 pandemic a few years ago made keeping these roles adequately staffed extra challenging, and the industry has been slow to recover over time. 

Nevertheless, today’s competitive food production industry has brought mounting pressures for companies to enhance food safety and product quality, achieve higher production throughputs, reduce waste, and operate efficiently. Given these demands, food production companies have few options but to evolve their approach to quality assurance and inspection. 

Food product inspection has come a long way

Many processing plants perform manual at-line product checks as their main method to assess process and product control. Let us use a hamburger bun manufacturer as an example. Quality control checkers observing the line may take a selection of hamburger buns off the line and walk them to a quality checking station several times a day. At the station, the QC checker will typically measure the bun size, shape, and thickness, often with a caliper or ruler, then weigh the product, and then check the product color using a reference photo. 

As one can imagine, this is a subjective and time-consuming quality control method. In the time it would take to evaluate even 10 hamburger buns, several hundred out-of-spec buns could have already moved through the packaging line. 

Forward-thinking food production companies began pursuing better methods to inspect products, leading to the development of rule-based vision inspection technologies. Consisting of high-resolution 2D and 3D cameras and specialized lighting to capture product images. These images are then analyzed using sophisticated software to extract measurements and detect defects. The measurements for each product are compared to user set parameters (rules) to accept or reject the product. Early rule-based vision systems were typically installed over a production line to inspect final products for basic features like size, uniformity, shape, and color. Rule-based vision systems help food processors significantly increase the speed and objectivity of their quality evaluation methods. With the help of integrated rejection methods to remove out-of-spec products from the production line, they could streamline inspection tenfold.

Naturally, as food brands began to see the benefits of automated product inspection, they began to incorporate more measurements into their system, pushing rule-based vision technologies to their operational limits. Not only was it challenging to define how to measure all of the desired attributes using a rules-based inspection system, it was also difficult for operators and QA personnel to set the more complicated parameters and understand the data from the vision system. Enter AI. 

Today’s AI-based vision inspection systems are trained using captured images labeled with information about desired product quality, defects, common foreign materials, and other relevant characteristics. Whereas rule-based systems are programmed only to inspect a handful of specific product features, AI-based systems learn using product images to distinguish individual features and defects with remarkable reliability. Technological advancements in vision system components like ultra-high-resolution camera resolutions, specialized lighting, and other hardware have accelerated the effectiveness of AI inspection applications. 

AI-based vision inspection technology emerges as a transformative solution, offering unprecedented precision, consistency, and automation – but often at a significant time and capital investment. The primary motivator for adopting these technologies is their clear potential to deliver a substantial return on investment (ROI) quickly, but how are food brands achieving this success? 

ROI driver #1: Avoiding product recalls from undetected foreign materials

Product recalls from foreign materials entering the process stream can harm a food brand. Each recall can cost several million dollars; only a small portion of that sum is the effort involved in removing the product from the market and distribution centers. Indirect costs such as lawsuits, lost sales, increased insurance premiums, and the public-relations effort to regain brand trust can be significant and impact the business profitability over several years.

Foreign materials can enter a food processing stream in several ways, from using raw materials with impurities—whether due to improper storage or handling of ingredients—to production equipment wear and tear. Today’s complex food production processes offer endless opportunities for foreign material contamination. However, human error or inattentiveness can also be a significant enabler of foreign material entry. Even as processing plants expand to meet increasing product throughput demands, expecting a small team of quality assurance personnel and inspectors to spot and remove foreign materials at high line speeds is unrealistic and unreliable at best.

A well-trained AI-based vision inspection system can detect potentially harmful foreign materials with high accuracy, often in places a human inspector cannot access. An AI-based vision system never leaves its post, never gets distracted, and never takes a sick day. And because of its ability to differentiate foreign materials from seemingly natural product features, it is especially effective for detecting soft, low-density objects like paper, foil, rubber, wood, colored plastics, and similar objects that would go unnoticed by X-ray or metal detectors. 

Considering the cost of a recall, if an AI-based vision system can spot and remove even one of these materials from the processing line, the system has already paid for itself several times over. 

ROI driver #2: Labor cost savings

As mentioned earlier, product inspectors typically have a high turnover rate in most food processing plants. The employee’s salary and benefits are only part of the total investment; the time and effort required to train an operator in their role also comes with costs. 

For food processing plant owners frustrated with the revolving door of QC checkers and similar other roles, a well-trained AI-based vision inspection can generate quick payback through fewer errors, reduced waste, and other production benefits. Some companies have been able to pay for their vision inspection system in under one year on labor cost savings alone. 

Also, integrating AI-based vision systems can allow companies to reallocate the laborers they do have from repetitive inspection tasks to more important and satisfying responsibilities in the plant. 

ROI driver #3: Superior process control saves waste

Rising ingredients and energy costs add another layer of pressure in food manufacturing facilities. With the help of AI-based vision inspection systems, which can help uncover adverse trends in process performance, food processors have a gateway to detect production issues before they become bigger problems. 

For instance, one cookie producer operating six production lines had a manual inspection protocol where personnel removed sample cookies after cooling every 20 minutes. About 25% of the total quantity of rejected goods was due to problems in the manufacturing processes. Nearly one-third of that value (9.1%) was explicitly due to the incorrect baking temperature. These routine errors amount to around 40,600 kg of wasted product over six months.

With real-time, 100% monitoring of products exiting the oven, the company’s production team could react much quicker and adjust oven temperatures as variations occurred throughout the day. Through this application, the cookie company reduced scrap waste by 8.7% (the remaining 0.4% of scraps occurred during the system integration process), amounting to a total of 38,800 kg in saved material.

Based on these figures, with an average cookie cost of $1.22/kg, these savings translated to nearly $47K in six months (38,300 x $1.22 = $47,336) and an annual savings of $94.6K. These savings only account for the reduction of wasted products; having the ability to make data-driven decisions on oven temperature also helped the company save energy costs.

It is also possible to use the inspection system data with the power of AI to help automatically control different parts of the production process. In the example above, it is possible for AI inspection systems to interface with oven controls and automatically control oven set points to maintain bake color of the cookies exiting the oven. This approach of using vision to control the product process is called Vision Process Control (VPC). 

Inspect less, earn more

Several food companies have found that integrating an AI-based vision inspection system can significantly enhance food production efficiency, quality, and profitability. When well-trained and maintained, AI-based vision systems provide precise, reliable, and continuous defect detection, dramatically reducing costly product waste and minimizing the risk of product recalls from unwanted foreign materials. 

By automating tedious inspection processes, floor workers at the plant can refocus on more strategic tasks, improving overall plant productivity. 

While the upfront investment may be significant, the measurable returns can swiftly outweigh initial expenses. 

Andrew has over 35 years of experience in the bakery industry and has worked for some of the world’s largest baking companies and food brands. Additionally, Andrew has over 10 years of experience in the development of automated vision-based quality inspection and food safety applications for baking operations.