Manufacturing Management with Big Data Enabled Key Performance Indicators (KPIS)

Manufacturing sector has conventionally been considered as the backbone of economic development and impacts the social, environmental and political well-being of any nation. Over the course of last two centuries, this industry has evolved by leaps and bounds, thanks to scientific advances, globalization effect and a growing world population. Multitude of factors such as raw material availability, labour costs, automation, logistics and government policies contribute to the dynamics of a manufacturing setup and its profitability. However, with sheer technological and managerial innovations, players such as Toyota successfully transformed traditional production principles and also sparked further research and exploration in this area. With their integrated socio-technical manufacturing system called ‘Toyota Production Systems’ (TPS), Japan achieved higher productivity and lower operation costs despite lacking natural resources and having geographical disadvantage in comparison to its American and European counterpart.

Manufacturing Management Overview

‘Toyota Production Systems’ comprises of techniques such as KANBAN (inventory scheduling), JIDAKA (quality control), ISHAKAWA (root cause analysis) and MUDA (waste reduction). In all these methods various parameters like cycle-times, wastes, stock-counts, demand, forecast, and machine breakdowns are logged and used for optimization analysis. These are crucial indicators which lead the management towards improvement goals and diagnostics. For example, to identify MUDA (non-value adding wastes) time-motion study of a process is done. Similarly, assessments of weight, time, temperature, noise, light, quality, etc. are mandatory for these improvement methods to be applied. These measurements were generally expensive and data intensive before the digital age.

To stay profitable and relevant in the competitive market of today, manufacturers must maintain high quality standards; low productions costs and optimize their processes by adapting smart manufacturing (SM) methods. The current wave of digital transformations offers ground-breaking opportunities with deep insights, which were untapped and invisible before. What separates the buzz word ‘Big Data analytics’ from other analytical methods are the 5 V’s of data, namely ‘Volume, Variety, Velocity, Veracity and Value’. Large manufacturing enterprises are sitting on huge stacks of data generated every moment by their 5 M’s (Manpower, Materials, Machines, Methods, and Money). Latest expeditions into Big Data, Artificial intelligence, availability of high-end computing power and low sensor costs has enabled the development of ‘data, information and knowledge systems (KS)’. Synergy of knowledge systems across different enterprise domains (Production, Maintenance, Finance, HR, supply chain) to achieve targets, company goals or corporate social responsibility is known as ‘Manufacturing Intelligence’ (MI). This paper explores MI architecture, various types of manufacturing KPI’s, methods to capture them and their applications in SM.

Key Performance Indicators: Before proceeding further, it’s noteworthy to delineate the difference between ‘indicators’, ‘performance indicators’ and ‘key point indicators’. An indicator may be any measurement of a quantity in the business like total shift hours in a month (shift Hrs. ) Indicators are generally meaningless until used in some context or contrast known as ‘benchmarking’. A performance indicator is a relative term which tracks an indicator value against a set value or target value. The number of ‘shift hours in a month per total hours in a month’ is a performance indicator of the plant utilization. (Shift hrs/hrs in month) There could be countless performance indicators which may or may not be of any significance to the management and may be carefully ignored. Key performance indicators are those critical performance indicators which measures ‘how much’ has been achieved against the target to ‘add business value’. KPIs are at the heart of any system of performance measurement and target setting. When properly used, they are one of the most powerful management tools available to a business. KPIs provide a definitive measure for production and resources performance. So continuing from the previous trail of examples, KPI would be the number of ‘productive’ shift hours per total hours in a month.

KPIs must have certain characteristics as expounded in:

1. Accountability: Each KPI should have a team or manager answerable to measure and work on its outcome.
2. Assimilation: They should be quantifiable, accurate & understandable by the shop floor and top floor.
3. Timely: KPIs should be measured frequently, reflecting current priorities. In smart manufacturing the KPIs are a real-time measurement.
4. Relevant: The measures should support strategic organizational objectives and add value.
5. Consistent: KPIs should not conflict with other performance measures. KPI’s are costly to capture and analyse hence only the most relevant ones must be selected. As mentioned before it is the prerogative of management to choose the most relevant performance measures. In figure 4 most commonly used performance measures by number are presented. 251 peer reviewed publications between the years 1979-2009 were researched to present this PI data pertaining to maintenance operations which accounts to 20% of total production costs. It can be observed that ‘cost’ and ‘OEE (overall equipment effectiveness) are the most frequent performance indicators.

MI Architecture: Henceforth we shall use the term MI (Manufacturing Intelligence to describe any knowledge system which gathers and analysis big data from the 5M’s to optimize and achieve business goals. It can enhance “Supervisory Control and Data Acquisition” (SCADA) by adding middleware or ‘smart substations’. ‘Smart’ in this context refers to the ability to connect with many different types of equipment using wired and wireless networks. It could have components like a microcontroller or may be a pure software application. Depending upon the requirements it is designed to provide a flexible interface with disparate control systems and data gathering abilities. Open standards like MTconnect agent for numerical controlled machines (CNC) is a good example of such a system. It is a protocol which gathers data from the NC machine through ‘adapters’ and streams it to an ‘agent’ in XML format. It has been widely adopted by the manufacturers.

The key functions of MI:

• Aggregation: ability to stream data through multiple sources.
• Contextualization: a structured and organized data model such as ISA-95 standard.
• Analysis: enable data analysis across various domains.
• Visualization: possess tools such as dashboards to make data comprehendible to decision makers.
• Propagation: Automate the transfer of data from the plant-floor up to ERPs such as SAP or Oracle.

In the following sections we would look at an implementation model of an MI system.

Implementation

This section describes a prototype to demonstrate implementation of an MI system based KPIs. It shows the data analytics approach for energy prediction of a CNC machine. As evident from the figure 6, the analytics platform is a symbiosis of data warehouse, analytics centre, physical shop floor streaming data via protocols, sensors and finally a virtual shop floor bearing the dashboard. In this particular use case following components were used: Mori Seiki NVD 1500 DCG for a machine tool, Fanuc 0i for a CNC, System Insights High Speed for a power meter, Cold Finish Mild Steel 1018 (10. 16 × 10. 16 × 1. 27 cm) for a work piece, and a solid carbide flat end mill (8-mm diameter, four flutes) for cutting tool. An MTConnect agent streams the data which is stored in XML files on a server.

The system design could be represented in following steps:

1. Modelling: As the goal is to predict energy consumption of CNC machine from its ‘program’ hence following function can represent the relationship between physical parameters of machine and energy consumption. Energy = f (federate, spindle speed, cutting depth, cutting width)+ error
2. Data Gathering: Value of these factors are acquired from the CNC through the MTConnect agent. NC programs are used to acquire the other data attributes, including command and trajectory after running.
3. Data pre-processing: The upper and lower 0. 5% values of energy readings are removed assuming they are outliers.
4. Synchronization: MT Connect documents, NC program data are synchronized with timestamps. Each block of the program is associated with the corresponding operation on the machine.
5. Dataset is divided into training and testing sets. 51 component models are used to train an ANN (two hidden layers and five neurons/layer) is applied using KNIME, a statistics and data mining tool. An artificial neuron network (ANN) is a computational model based on the structure and functions of biological neural networks. Information that flows through the network affects the structure of the ANN because a neural network changes - or learns, in a sense - based on that input and output. ANNs are considered nonlinear statistical data modelling tools where the complex relationships between inputs and outputs are modelled or patterns are found.
6. Computation and Model Validation is done on 12 of the trials and energy consumption was successfully predicted less than 1% total relative error (TRE) with the measured energy. Results are given below.

Conclusions

This paper presented the value, design and implementation of KPI based smart manufacturing using data analysis to achieve advance decision making capabilities and production optimization. The system could also facilitate development of digital twins and seamless integration of the physical and virtual process. Industry agnostic KPI’s like OEE (Overall equipment effectiveness) can be measured in real-time by mapping the process bottleneck to the data models. Open standards such as MT Connect are a welcome step towards IOT 4. 0 realization and the future belongs to interoperable, scalable, collaborative and predictive ‘knowledge systems’ (mentioned in intro section) for smart manufacturing management.