Smart Manufacturing

Introduction

Smart Manufacturing, or Industry 4.0, creates many opportunities for enhancing your products and operations by using Artificial Intelligence (AI) technologies within the smart manufacturing movement, converting your operations into a “smart factory”. Once the domain of “Early Adopters”, AI technologies in manufacturing are now mainstream. In the process industries, these technologies enhance your ability to understand cause and effect in product performance, model, predict and optimize operations to enhance yields, reduce rework, and use more affordable raw materials with greater confidence. In discrete manufacturing, AI algorithms identify optimal combinations of subassemblies that can be used to ensure the best combination of subassemblies across the entire inventory and create uniform, consistently performing products. These AI algorithms can also ‘dial in’ the necessary product settings to achieve desired performance measures. This ensures that all assembled products meet strict customer requirements and standards. Since all products are made on-spec the first time, rework is often reduced to zero, greatly improving manufacturing performance and increasing capacity without a capital investment.

Artificial Intelligence (AI) brings a new dimension to the manufacturing industry by enhancing materials, process and product data management, machine learning, performance estimation, performance prediction, failure analysis, and a broad array of optimization technologies and approaches. This enables manufacturers to enhance conformance to customer requirements, reduce rework, increase utilization, and lower costs significantly. AI modeling, prediction, and optimization benefit the entire manufacturing industry.

Definition of smart manufacturing

Smart Manufacturing, or Industry 4.0, combines advanced AI technologies and optimization processes to improve efficiency in manufacturing operations. Manufacturing can generally be divided into two broad categories, Process Industries and Discrete Manufacturing. Process Industries involve continuous processes such as oil and gas, refining, chemical, pulp, and paper. Discrete Manufacturing often involves assembling and packaging. Most manufacturing has elements of both.

Discrete assembly operations data is often batch or lot oriented and more transactional. AI algorithms such as machine learning, predictive analytics, and genetic algorithms for combinatorial search are used to identify optimal combinations of subassemblies. The objective is to assemble subassemblies so that all produced products conform to specifications, a global optimization, rather than finding the best of the best single combinations that result in unusable leftover subassemblies.

In the process industries, the data is often time-series, expressing the flow of raw materials properties, processing conditions, and product performance through time. However, some batch-oriented operations, such as mixing tanks of raw materials, milling, and the like, are often found. AI algorithms such as machine learning, predictive analytics, continuous gradient optimization, and “Model Predictive Control” are used to identify, predict and optimize based on the causal relationships between materials, process conditions, and quality results.

Modeling, Understanding, Prediction, and Optimization Through AI

AI modeling algorithms, sometimes called “Machine Learning” or “Deep Learning”, identify and predict product performance using the causal relationships between process conditions, materials properties, and product performance. These models can be explored graphically or through “Sensitivity Analysis”, that is, to run simulations to understand better which factors drive product performance and which interact with each other. The expected performance can be estimated by running current conditions and materials characteristics through the models. If these estimates are forward-looking in time, they are called predictions.

Once these relationships are understood, AI-powered optimization algorithms can be used to control inputs to maximize performance results. For example, in “Model Predictive Control” (MPC), current uncontrollable inputs are supplied to the models, and the models are searched for controllable inputs that result in a desired product performance. The resulting controllable inputs can be used as setpoints in Distributed Control Systems (DCS). Optimization is often based on global objectives such as targeting a specification, maximizing yield, or minimizing costs while meeting hard limit constraints on setpoints and product quality standards. Optimization technologies are used in production systems, from single machines to entire plant networks, within and across smart factories.

Please note that cloud-based “big data” systems that provide big data processing capabilities are not necessary, nor are “digital twins” (simulators), or “data lakes”. We rely on local databases and your on-premises industrial internet to convert traditional manufacturing processes into smart manufacturing. We provide emerging technologies to complement your manufacturing systems with data analytics, modeling, prediction, and optimization to deliver a competitive advantage.

Outcomes of Optimization

Optimized operations reduce rework, improve productivity, reduce energy costs, increase yields and throughputs, and use more affordable raw materials more effectively for consistent product performance at higher profit margins.

For one customer in aerospace, we were able to reduce rework from 60% to zero. Yes, ZERO. We modeled the effect of properties of multiple types of subassemblies vs. over a dozen product performance metrics, then took the entire inventory of multiple assemblies and did a genetic algorithm AI search, virtually assembling them, estimating performance, and then issuing matched pick lists that resulted in ALL assemblies meeting specification. The system was so successful the customer received an industry award for improvement.

In pulp and paper, we modeled product performance vs. various costs and grades of pulp and determined the best blend to achieve desired properties at the least cost, resulting in savings of $185,000 per month, greater than 300% ROI per month (not per year!).

In oil and gas, we estimate production and artificial production assistance and determine accurate rate estimates and how to use production assistance to maximize production and optimize downstream network constraints so the maximum amount of oil, gas, and water are delivered within the capacity of the central processing facilities.

Conclusion

Using Artificial Intelligence (AI) to create smart manufacturing is a powerful method to maximize performance and reduce costs and is becoming the norm for doing business competitively. With AI modeling and optimization algorithms, process engineers can model the effect of different inputs on product properties and optimize setpoints to meet product specifications at a reduced cost. Optimizing operations with AI-powered optimization algorithms increases yields, reduces rework, saves energy costs, and improves productivity while meeting hard limit constraints on setpoints. The result is consistent product performance that delivers higher profit margins for manufacturers.