By Arvind Saraf
There’s an old saying that goes, “You can’t improve what you can’t measure.” That’s why AI is making such an impact on data-heavy industries. But should AI be used to measure machines or humans? Let’s take a close look at one industry, manufacturing, where machines have been extensively measured for years, while humans have not been measured effectively.
While there is an abundant discourse about how artificial intelligence is transforming the way manufacturers source data and analytics to understand and address some of the biggest issues they have been facing when it comes to machines, very little has been discussed about a large pool of data that has been inaccessible for more than a century– actions performed by humans.
Despite all the hype surrounding automation and robotics in manufacturing, the simple fact is that humans still do the bulk of labour on assembly lines. A closer look at the numbers clearly indicates that it will take at least a 100 years before automation is more abundant than humans in manufacturing. A recent survey shows that up to 72 per cent of manufacturing tasks are still performed by humans and 71 percent of the value created by the operation comes from human actions. This figure is not surprising when you consider the cognitive ability, adaptability and dexterity humans bring to the table.
However, human activity on manufacturing lines has effectively remained invisible to analytics. Trying to harness this invaluable data, let alone put it to good use, has long been a challenge. Unlocking this data becomes all the more significant when you realise that almost 340 million humans, (according to a Goldman Sachs Research,) are still employed across various plants. This data further lends weight to the argument that human performance analytics is a clear requirement for success in manufacturing for the foreseeable future.
The limitations of time and motion studies
The conventional methods to monitor human output using stopwatches or physical observation are fraught with several limitations. This process is time-consuming, labor-intensive, susceptible to observational bias, does not record behavioural data accurately and can lead to incomplete documentation and analysis. Besides, these processes can only measure the length of time it takes to complete the process and fail to observe other variables that need measuring, such as process adherence and output quality. Such limitations make it difficult for companies to make accurate assessments relevant to the performance and efficiency of their people and processes followed — and therefore make smart decisions about staffing, production, training and more.
Adding to the challenge is the sheer complexity of human actions. This is especially true for manufacturing. Although it seems like a simple task, to mount a circuit board, for example, there are several steps involved, making the process vulnerable to variability. The aforementioned survey points out that 73 percent of the variability on factory floors comes from human workers and not machines. While a robot can be programmed to do a job a thousand times without variation, the human output will vary in terms of quality, duration taken and more. The output could also differ from person to person, as no two people are the same.
AI provides extensive data — and insights —on human actions
So, how can manufacturers address this issue? As it stands now, video is the best way to capture this information in raw format. And thanks to breakthrough innovations in advanced data analytics techniques, such as computer vision and machine learning, manufacturers now have the ability to gather such footage in real-time, as well as associated data, which can be shared across the board from line workers to plant managers to improve efficiency. This has not only made huge impacts on process optimisation but has opened new avenues for the acquisition of data and meaningful insights on manual assembly.
AI at work in the factory
Standardised work is the gold standard in manufacturing: it’s the way to complete a process that optimizes efficiency and quality. Many manufacturers struggle to ensure standardised work is being followed on every station, through every shift. That’s where AI plays a key role. Advanced computer vision algorithms can be used to validate and measure critical elements of manual production lines using proprietary action recognition technology. This technology learns a line’s standardised work instructions and observes workers as they perform steps, identifying any deviations in the manufacturing process like missing a key step or an out-of-sequence action by their workers. In some instances the line associate is notified right away; in others, the video is used for training purposes. This helps minimise production defects.
Here’s another example where AI can observe humans to benefit manufacturers. Every discrete assembly line experiences bottlenecks, where one station is slower than the others and lowers productivity across the line. Today, manufacturers use time and motion studies or simple observation to determine the cause of bottlenecks, but it is very imprecise. For example, on an assembly line, it may look like station three is the bottleneck, but a materials shortage at station two may be the real culprit, which isn’t fully evident at a glance. However, a consolidated view of objective data for the entire line makes these hidden causes much easier to discern. With this powerful information, manufacturers will have the ability to make important decisions like providing additional training to the line associate at station two or improving material flow, so as to reduce the downstream delay. AI allows continuous automated measuring of this data over time, allowing to track any operator-specific areas of improvement.
So, to answer the question – “What should AI measure, humans or machines?” The two are not mutually exclusive. It’s important for manufacturers to have a clear picture of what’s happening with the machines in their factory, but it’s also critical to address that human-shaped blind spot that has existed on the assembly line for over a century. And to do that adequately, AI is an excellent tool because of its ability to process every action, turn it into data and present clear analytics and insights to the manufacturers when and where they need them most.