Automated Quality Control in Manufacturing

Artificial intelligence is transforming the face of nearly every industry on the planet. Consulting firm McKinsey & Company estimates that in 2016, businesses poured more than $26 billion into investments in AI. By 2035, AI technologies may be able to improve productivity by 40 percent and provide an economic boost of $14 trillion across 16 different sectors.

The first step of AI is to teach the machines. This involves training computers to find patterns in massive data sets and then using those observations to make smarter decisions, often referred as machine learning.

In particular, machine learning is commonly applied to computer vision. By analyzing large sets of picture-based data, computers can learn to do everything from recognizing handwritten digits to describing the objects in a photograph. Thanks to a breakthrough at Google Research in 2013, computers are able to recognize 100,000 different types of objects within an image after only a few minutes of analysis.

Andrew Ng, renowned AI expert and former head of the Google Brain AI research project, recently started the company to focus on the applications of AI in manufacturing. The manufacturing industry currently faces a number of challenges, such as unexpected variations in product quality and quantity, the inability to rapidly scale and manage capacity, and the rising costs of production. Fortunately, help is on the horizon.

Before they leave the production line, items must be inspected in order to make sure they meet quality and safety standards. Today, these inspections usually require intensive human labor in order to visually examine objects and identify problems.

The arrival of machine learning and computer vision techniques on the production line will be nothing short of a revolution for manufacturing. The field of computer vision is becoming more and sophisticated, and machines can now do far more than they could even a few years ago. Machine learning researchers at Google are working on everything from robot motion planning to real-time image enhancement.

These advances in classification and understanding complex imagery are on the cutting edge of research in machine learning and computer vision. Although this level of expertise has been around for several years, up until now it’s been largely confined to major enterprises with deep pockets.

However, these barriers and restrictions are rapidly disappearing. Thanks to the release of open-source software libraries for machine learning, such as Google’s TensorFlow, image analysis and object detection are becoming easier and more mainstream.

By working with experts in machine learning and computer vision, manufacturing companies will be able to leverage these developments in order to improve their productivity, efficiency and revenue.

Packaging, for example, is an extremely important step in the manufacturing process; it’s used for branding and product identification but also for logistical purposes such as shipping. By imprinting each package with a unique code, the manufacturer can keep track of its timing, origin and routing. In so doing, buyers will be better informed about how long shipping will take and when their package will arrive.

Unfortunately, anomalies in the packaging process can result in costly errors and even product recalls. For example, unlabeled allergens such as milk and peanuts have been the source of roughly half of food product recalls. Other sources of recalls include applying the wrong label and including an incorrect translation on the product’s packaging.

With an average of $10 million in direct costs alone, food product recalls can be devastating to a company’s profitability. All too often, these problems are caused by packaging and labeling mistakes and other human errors.

By visually inspecting items, AI-enhanced computer systems can automatically determine the correct labels that should be applied to the product, slashing the potential for errors. Machine learning therefore helps ensure that customers receive the right product and that the product isn’t lost or delayed in shipping by having an incorrect label.

Modern, innovative solutions such as SpringML’s label detection app can take advantage of machine learning techniques to reduce errors in package labeling. With speed a critical factor, you need a solution that can extract text in real time from photos and videos — and raise an alert if it detects an anomaly. Any such solution must be able to handle a variety of labels and input images as well as variations in lighting, camera angle and image quality.

Of course, quality control and visual inspection are just the tip of the iceberg when it comes to the potential uses of machine learning in manufacturing. For example, AI-enhanced computers can now analyze sensor data on the different parameters of manufacturing equipment, such as temperature and pressure. The combination of these parameters can then be used to predict when each piece of equipment will require maintenance, helping to reduce lost productivity due to unexpected breakdowns.

With so many potential applications, it’s little wonder that market intelligence firm TrendForce estimates that the “smart manufacturing” market will grow to $320 billion by 2020.

Machines excel at repetitive tasks like the ones that manufacturing companies require. By training computers to visually inspect products on the assembly line, humans can reap the rewards of reduced errors and faster rates of production.