In today’s volatile, uncertain, complex and ambiguous (VUCA) world, an agile and cost-effective manufacturer needs to cater to market demand timely and with quality products. This creates the need for manufacturing sys­tems to be pull and precision-driven.

Historically manufacturing has been rooted in observing past data, analysing the same, deciding how to improve and then act­ing on the conclusions. Machine learning not only allows observation of historical data, but forecasts demand, analyses past and future in tandem to take decisions that will pre-empt glitches, and ultimately enables businesses to be proactive (over-reactive), cost-effective (over cost controlling) and lean.

Machine learning is iterative, causing al­gorithms to learn with each repetition and optimise actions to be taken. Deep learning, a further refinement of machine learning, is based on pattern recognition. It allows algo­rithms to identify outcomes by superimpos­ing past patterns on a new real-time one.

Being measurable, machine learning can easily be used to change old processes and execution design. World leaders such as Sie­mens and GE are investing billions of dollars to transform their manufacturing setups. It is not long before that even smaller firms will follow suit.

This article discusses the vision and chal­lenges for adopting machine learning in core manufacturing and supply chain processes. Machine learning for product design de­serves standalone attention.

Organisation Vision for Adopting Ma­chine Learning

Enhance Supply Chain

• Forecast Demand &analyse past

customer purchases

• Control inventory levels

• Optimise time to market

Improve Production Efficiency

• Govern quality levels& reduce

wastage

• Optimise batch sequencing

• Optimise MRO times

Enhance Competitiveness

• Identify underserved markets & competitive advantages

• Enhance product pricing in real time

• Create leaner decision making

Machine learning can be effectively used to optimise supply chain with ac­curacy and precision like never seen before. Modeling different customer requirements and purchase patterns, the need for a product can be under­stood; consequently, calculating raw material requirement. Extrapolating future requirements allows smoother supplier-buyer interactions and WIP (work in progress) build up. Maintain­ing inventory at optimal levels ensures that lack of material does not create a loss of sales. This can be vital if mate­rial procurement has a long lead cycle or the just the availability of the right WIP can reduce assembly times. With the iterative and forecasting ability of machine learning, not only will work­ing capital management be simpler, but more importantly supply chain processes will become transparent.

“The future of manufacturing with machine learning, capabilities of which are limitless is exciting”

Algorithms allow predictably to be built into operations. Sequencing batches correctly can ensure the most efficient utilisation of bottleneck and critical machines and optimal alloca­tion of manpower. Forecasting main­tenance requirements based on usage patterns and data from machine parts can allow timely predictive mainte­nance and lowest downtime for pre­ventive maintenance. In fact, stock out of critical spares or over stocking can also be avoided. Using sensors, machine data collection, and pattern recognition; machine learning can help in stringent monitoring of qual­ity parameters -reducing wastage and increasing first-time right production. The immense power of algorithms can add generously to the profitability of manufacturers.

VUCA world calls for organisa­tions to be one step ahead of peers. Deep learning facilitates expedited recognition of markets that have a need for a product but either are un­derserved or have room for up-selling. Working with data sets of customer sales, patterns of cross-selling may emerge that can help capture market share, increase revenue and even adopt omnichannel strategies. Deal analytics can allow for better matching of cus­tomers to products, and dynamic and transparent pricing. Both these create better customer experience and sales outcomes, enhancing brand value.

Challenges in Adoption of Machine Learning

• Understanding Data Requirements

• Understanding Infrastructure

Requirements

• Understanding Talent

Requirements

Applicability of machine learn­ing from supply chain to production to sales to contract management, and its ability to improve the efficiency of every process makes its Return on In­vestment a no-brainer. Yet there are barriers to adoption.

The first hurdle to be crossed con­cerns data organisation and security. While most manufacturing organisa­tions have terabytes of data, most of it is unstructured and/or fragmented making data unusable. Consolidating and structuring data is a time consum­ing laborious exercise. This bland and uphill challenge needs to be overcome by every CIO to reap the resulting multifold benefits.

Data storage and data security are the next decision-making fac­tors. To fully harness the power of machine learning, it is imperative that cloud storage and/or hybrid op­tions be considered. There is no one size fits all solution. CIOs will need to weigh organisation maturity and strategy to find the best alternative for their firms. With large data and a diverse network, analysis of perfor­mance and security is critical, as is an environment that makes testing and experimentation unproblematic.

However, the third and the most vital consideration is that of talent. Not only are competent data scien­tists and good developers needed, but there is also a need for trainers to help business adopt machine learning in their daily life.

The future of manufacturing with machine learning, capabilities of which are limitless is exciting. It is up to the leadership to visualise, lev­erage and implement the technology to create a more agile, profitable and leaner company.