Today is World Logic Day, which is an opportunity to bring attention to the history, significance, and practical implications of logic.
Reminding ourselves of rational decision-making is especially relevant today – with heightened election claim disputes, government propaganda, social media trolls, and contentious sporting decisions.
For many, the result of any logical assessment is one of two extremes, bivalent: true or false. Although the simplicity of this approach allows it to be consistently implemented in the multitude of semiconductor-based devices, it artificially excludes the subtleties necessary in real-world situations.
I was introduced to fuzzy logic during the 1990’s, when it was being increasingly implemented in industrial control systems. Fuzzy logic was proposed by logician Lotfi Zadeh and is a form of many-valued logic where the truth value may be any real number between 0 and 1, making it well suited to handle partial truths.
Fuzzy logic based industrial control systems are particularly suited to situations with high uncertainty and nonlinear behavior, and there are no precise mathematical models available. Its first notable application was on the Sendai Subway Namboku Line, Japan in 1987, and has been used in air conditioners, camera auto-focus systems, refrigerators, combustion engines and robotics since then.
PLCs (programmable logic controllers) are the decision engines in industrial equipment and processes. These devices have enabled considerable advancements in automation, however their historical reliance on specific hardware has limited their use in fuzzy logic scenarios due to the mathematical processing required.
The growing proliferation of software-based PLCs, in which the control logic is decoupled from the hardware now allows complex logic to be executed in the applications they run. This enables more equipment to benefit from the reliability improvements, reduced wear and tear, and energy efficiencies from fuzzy logic control, and opens the door to incorporating AI into industrial process control.
AI and logic in industrial IoT
In the industrial IoT sector, AI is already being used to deliver a wide range of benefits, which typically fall into three groups:
- Real-time assessment of multiple streams of time-series data like temperature, pressure, and axis position to determine trends and predict failures from known fingerprints, like paint shop robots proactively correcting operational faults
- Assessment of the quality of manufactured products from automated visual inspection and pattern detection, like the automated quality control in high volume textile manufacturing
- Assessment of both trends and outliers for anomaly detection, which allows a unique digital model of the normal operation of each asset to be created, that can then be used to identify abnormal operation, like in the efficient operations of next generation wind turbines
Although the benefits of AI for industrial IoT uses is becoming clearer there are still some challenges in its execution. The shortfall of data science competence is one: There are three times the number of data scientist job postings versus other job searches. Second, the exploitation of this competence is extremely low, with 87% of data scientist time spent on non-data science tasks.
To exploit the benefits of AI there are three fundamental capabilities required:
- First, I believe that the power of AI should be put in the hands of the domain experts through self-service tooling. Allowing factory floor technical engineers to self-create AI models that leverage their expertise and experience using no-code tools
- Second, as the early AI pioneers have already discovered, the lifecycle management process of AI needs to be hyper-automated with machine learning (ML) operations tools, which allow the valuable data science resources to be focused on data science tasks
- Third, it is fundamentally limiting for any enterprise to restrict the automated decision-making capability of AI to a single location in the architecture, due to issues like business continuity, local autonomy, access to data, and latency. Enterprises that can create AI / ML models in one place and deploy anywhere, across cloud, core and edge, without any alteration of the model AND any alteration of the execution of the model, are able to exploit AI / ML wherever they need.
It is only through providing these three fundamental capabilities can enterprises exploit the immense potential of AI, and fuzzy logic, in their industrial IoT deployments.
Happy World Logic Day!