Elsevier

Journal of Cleaner Production

Volume 137, 20 November 2016, Pages 1628-1646
Journal of Cleaner Production

Measuring variability on electrical power demands in manufacturing operations

https://doi.org/10.1016/j.jclepro.2016.03.102Get rights and content

Highlights

  • A method for measuring the variability on manufacturing power demands is proposed.

  • The manufacturing peak power demand is estimated by using the variability.

  • In a case study, the variability and peak demand are estimated with acceptable errors.

Abstract

Manufacturing energy studies have generally focused on estimating the mean electrical power demanded by machines. Variability is, however, also an important factor to consider in manufacturing power studies, since an understanding of electrical demand uncertainty can support the estimation of peak power demand, which impacts energy costs and electricity distribution systems. For example, U.S. federal law requires power suppliers to provide credits to large electricity consumers, such as manufacturers, when they enter into pre-established peak demand reduction agreements. The peak demand in a manufacturing plant can also be used to plan and design a factory electrical system, and to determine the amount of capital to be invested in facilities and equipment providing electricity services to new, large customers.

Thus, this paper proposes a systematic method for estimating the peak demand based on readily available information such as manufacturing parameters. More specifically, the proposed method extracts the power demand mean and variance from manufacturing parameters at the machine state level by considering various manufacturing processes. The machine state level mean and variance are then used to approximate the mean and variance at the machine level by implementing the probability mixture model. Applying the Lindeberg central limit theorem to the machine level information, we estimate the mean and variance of power demands at the manufacturing system level. Then, a method for estimating the peak power demand using the mean and variance at the system level is proposed.

In an illustrative example, we demonstrate our proposed method for real manufacturing power profiles, including milling, turning, and welding. We show how to connect manufacturing process parameters with the mean and variance of power demands and how to estimate the peak power demand based on the mean and variance. To validate the proposed method, we simulate a hypothetical manufacturing system; the results suggest that the proposed method can estimate the mean power demand with 2% error and standard deviation with 12% error, and that the peak power demand can be estimated with 20% error even in the worst case.

Introduction

Discrete manufacturing energy analysis is currently an actively researched topic, for two principle reasons: energy costs are expected to fluctuate; and energy-awareness has become more important in manufacturing. Studies of manufacturing energy can provide a tool to help manufacturers make more cost-effective and environmentally-friendly managerial decisions. While various types of energy are consumed in discrete manufacturing plants, most manufacturing machines require electricity as a main energy source. Thus, research on electrical energy consumption can cover the broader aspects of discrete manufacturing energy analysis.

When analyzing the electrical energy required by a manufacturing system, we must consider not only mean demand (MD) but also variability of demand. Energy consumption can be estimated using MD, and peak demand (PD) of electrical power can be calculated from MD and variability. MD and PD can be shown as in Fig. 1. If demand is considered a function of time (t), energy consumption is the area under the demand. MD is defined as energy consumption divided by a time period (t1), and PD is defined as the maximum demand value during t1.

While data on MD and PD of a manufacturing plant can be useful in various ways, one of the most significant applications is in estimating electricity costs for time-of-use (TOU) pricing. For example, the Energy Policy Act of 2005 passed by the U.S. Congress requires power suppliers to provide customers with TOU pricing rates upon customer request; large electricity consumers such as manufacturing plants in the U.S. can receive credits by entering into pre-established PD reduction agreements that reduce planned capacity obligations (U.S. Congress, 2005). MD and PD can also be used to calculate electricity costs directly. One U.S. power supplier (National Grid) introduced an electricity pricing plan charging not only for energy consumption but also for PD.

Research on estimating MD and PD can also provide input to other research areas. In particular, production rescheduling models for TOU electricity pricing require accurate MD and PD estimates as input parameters. Applying studies on production rescheduling, manufacturing plants can minimize electricity costs by keeping the total power demand below the allowed maximum value. Studies by Ashok (2006) and Babu and Ashok (2008) introduced relevant production rescheduling optimization models for a manufacturing plant, considering energy consumption and PD charges. Consideration of MD and PD will also be beneficial in the design of manufacturing plants, since research on MD and PD can provide analysis that is not possible using only MD. In the design stage, estimated PD can make useful contributions to the selection of both power supply and distribution equipment. Hence, in order to help manufacturing plants to reduce energy costs (and possibly energy use) and to offer crucial input to other relevant research areas, studies on electricity demands for manufacturing must consider both MD and PD.

Estimating MD and PD is, however, not straightforward, since the two demands are dependent on many variables. Any change in operation hours, worker shift schedule, types of machines, product design, raw material, or manufacturing processes may influence energy demand, creating hourly, seasonal and annual variation in MD and PD (Ashok, 2006, Thiede et al., 2013). The PD spike shown in Fig. 1 could be caused by any of a number of the operating decisions, and estimating MD and PD from operational parameters is difficult in general. In other words, a method to estimate MD and PD is urgently needed for existing and future manufacturers, but a systematic, unified method to connect managerial/operational decisions, manufacturing parameters, and MD and PD together is not currently available.

Section snippets

Background

Most of the conventional manufacturing systems are modeled and summarized by Altiok (1997). In this book, typical manufacturing systems, such as a job shop and flow shop are introduced, and probabilistic approaches to modeling the systems are discussed, with a focus on performance analysis. Jebaraj and Iniyan (2006) provided a broad review of the literature on energy models, including energy planning, energy supply-demand, and forecasting. This review deals with macro-economic types of energy

Modeling manufacturing power demand

In conventional manufacturing models, system performance, measured in terms of throughputs, machine utilization, flow time, and customer service levels, has been a major focus of research. In energy-aware manufacturing models, however, we consider additional factors, including the power demand and energy consumption of a manufacturing system. While energy-aware models depend on power and energy parameters, which have not been a part of the conventional models, we can reconcile the conventional

Illustrative examples

In Section 4, we discuss modeling of the power demands of manufacturing machines and how to apply the proposed models to estimate the system level peak demand and energy consumption. To that end, we analyze measured power profiles from real manufacturing machines, and propose methods to characterize each profile as manufacturing parameters. As discussed earlier, we discretize the time horizon; the time interval is 0.5 s, the finest sampling interval of the measuring device (Fluke 1735).

Results and discussion

Following the proposed approaches, we show the results of the case study, and discuss the significance of the results. More details on the case study results are provided in Table 15, in which we compare the simulated results (SIM) with the estimates by the proposed method (EST).

For each case, we estimate μW and σW by applying our proposed method. Among alternative methods, we use the designated method found in Subsection 4.1, including MTA (c = 1.44). After aggregating all Xi in each

Conclusions

This paper presents a systematic method to measure manufacturing power demand variability as MD and PD using manufacturing parameters at the machine state, machine, and system levels. Since the proposed model requires only readily available manufacturing parameters (process plans and machine utilization) as inputs, the proposed approach can be implemented with less effort and resources than a simulation-based method requiring machine power profiles. Our proposed method begins with models at the

References (37)

  • L. Suganthi et al.

    Energy models for demand forecasting—A review

    Renew. Sustain. Energy Rev.

    (2012)
  • S. Thiede et al.

    Environmental aspects in manufacturing system modelling and simulation—State of the art and research perspectives

    CIRP J. Manuf. Sci. Technol.

    (2013)
  • N. Weinert et al.

    Methodology for planning and operating energy-efficient production systems

    CIRP Ann. Manuf. Technol.

    (2011)
  • R.N. Allan et al.

    Probabilistic analysis of power flows

    Proc. Inst. Electr. Eng.

    (1974)
  • T. Altiok

    Performance Analysis of Manufacturing Systems

    (1997)
  • C.A. Babu et al.

    Peak load management in electrolytic process industries

    IEEE Trans. Power Syst.

    (2008)
  • B. Borkowska

    Probabilistic load flow

    IEEE Trans. Power Appar. Syst.

    (1974)
  • A. Cannata et al.

    Energy efficiency driven process analysis and optimization in discrete manufacturing

  • Cited by (11)

    • Risk-based optimal bidding patterns in the deregulated power market using extended Markowitz model

      2020, Energy
      Citation Excerpt :

      In a single-settlement market, competitors purchase or sell wholesale electricity during the operating day, and electricity prices are set daily or hourly depending on the demand and available supply. In a two-settlement market, competitors purchase or sell wholesale electricity one day before the operating day on the day-ahead electricity market, and the difference between the proposed and actual demand on the operation day will be covered using a real-time market [2,3]. In the day-ahead market, every generation company (GenCo) (or participants) submits a bidding pattern and strategies to maximize its profit by optimizing the bidding strategy [4–6] and to the Independent System Operator for each hour of the next day.

    • Estimating manufacturing electricity costs by simulating dependence between production parameters

      2019, Robotics and Computer-Integrated Manufacturing
      Citation Excerpt :

      This suggests that simulation models combined with the copula approach can be used for studying the dynamics between production parameters and electricity costs. Given the various manufacturing processes, more recent studies have begun to investigate how production parameters can affect manufacturing utility costs [9,10]. This implies that a manufacturing system with milling machines can be more readily analyzed for simulating the dynamics between utility costs and production parameters.

    • Power demand risk models on milling machines

      2017, Journal of Cleaner Production
      Citation Excerpt :

      Other studies have also presented power demand analyses of individual manufacturing processes. Studies by Avram and Xirouchakis (2011) and Jeon et al. (2016) proposed methods for reconstructing power profiles of a milling machine based on NC code and the coordinates of the machine spindle during processing. Specific energy consumption (SEC) of injection and compression molding processes was investigated by Krishnan et al. (2009) through power measurement experiments with six injection and four compression molding machines.

    View all citing articles on Scopus
    View full text