09 May Efficient Manufacturing
Artificial intelligence for an energy and resource efficient manufacturing chain design and operation
The energy and resource efficient manufacture of consumption and investment products is becoming a competitive advantage and companies are increasingly interested in optimal manufacturing chain design and process operation. Based on the discrete events modeling approach empirically parameterized process models for heating, hot-rolling, forging and turning are combined to two alternative manufacturing chains for the manufacture of countershafts. The discrete events also consider specific NC codes (e.g. for turning) and allow for time-depended consumption profile calculations. Further defining and structuring all parameters of the manufacturing chains with all their processes in so- called system entity structures provides the basis for a numerical optimization by artificial intelligence tools. A genetic algorithm in combination with a fitness function has been employed to find the manufacturing chain design and process parameter set with the lowest energy and resource consumption for the manufacture of the shafts in an effective way.
The improvement of the efficiency of machines and manufacturing systems as well as the reduction of manufacturing costs have been important since the beginnings of industrial manufacture. However because of ecological and economic reasons, the energy consumption and more general the utilization of resources in manufacturing are becoming increasingly important. A continuously rising energy demand, coming along with a shortage of energy resources, causes energy costs to rise and to increase the burden on the ecosystem as well, as the main sources for the world energy consumption are still the fossil fuels. Energy and resource efficient manufacture of consumption and investment products is becoming a competitive advantage and companies are increasingly interested to design their manufacturing chains in an optimal way.
There have been various analyses targeting on the reduction of e.g. the energy and resource consumption of machine components , the energy consumption and the total life-cycle costs of manufacturing systems or of single processes [2, 3, 4]. Helu et al.  discuss the surface quality of machined parts relative to energy, resource and service costs. Teti et al. presented advanced methods for monitoring process states . In order to allow for a flexible or alternative systems analysis regarding energy and resource efficiency as well as for its prediction during the design and planning phase of manufacturing processes as well as chains, capable modeling and simulation approaches are necessary such as described in [7, 8, 9]. Hagendorf has described such a method . He employed the so-called system entity structure approach, which describes and structures the relations between the entities of a system, and modelled the manufacturing process of a workpiece by discrete events of process states that take place during the operation of a tool machine (cf. figure 1). When a part enters the event steering block (here 2- axis turning), parallel or successive machine control events are initiated at a given time according to the NC code of the specific part. All temporary energy and resource consumption are integrated over time and all changing states (e.g. new shape) are recorded and returned to the entity, respectively the workpiece. Larek et al. [7, 11] have used the system entity structure and discrete event approach by Hagendorf to analyze and predict the energy consumption for a turning operation of a shaft. For this purpose all energy consuming entities and discrete events of a 2-axis turning machine respectively process were described and parameterized on basis of measurements. The comparison of the measured and calculated power consumption profile showed very good agreement.
In this contribution the approach by Hagendorf and Larek [7, 10] is extended to the manufacture of a simplified countershaft considering metal forming processes. To demonstrate its potential for a resource efficient manufacturing chain design two rather classical types of manufacturing chains were compared: a rather short cutting based one (turning) and a hot-metal forming based manufacturing chain. The analysis focused on the consumption of operation resources like materials (material removal, swarf, tool wear), fluids (drag out) and energy (electricity, gas).
The manufacturing process chains and process modeling
For the comparison of the two manufacturing process chains, it is necessary to define reasonable boundary conditions. In this analysis we chose to start from a bar stock material (Ø 62 mm) which allowed the direct turning of the shaft shape after cutting off cylindrical sections from the steel bar. Figure 2 shows the bar stock material data in the upper central block from which the two manufacturing chains are starting. They end at the turned sample shaft (center of figure 2).
The manufacturing chain involving hot-metal forming, starts with heating the bar stock material, followed by rolling it down to Ø 42 mm in order to make a better use of the material and to reduce the necessary material removal. Cylindrical sections are then die forged to the final shape, considering 2 mm for material removal. Following the considered shaping operations forming and turning possible toothing, heat treatment and grinding operations would follow for the manufacture of a countershaft. Since these processes would be the same for both manufacturing routes, they would contribute the same amount of resources consumption to both manufacturing chains analyzed here. Therefore they were neglected in this analysis.
In this analysis the so-called basis models for turning and heating developed by Hagendorf  and Larek et al.  were employed. Hence only the measurements, parameterization and process representation by discrete events for the additionally developed hot-metal forming processes, respectively the rolling will be discussed here. For the discrete event approach it is necessary to distinguish between a constant consumption level of a process, respectively machine, which is ready-to-operate and the load-dependent consumption state while in operation.
Figure 3 shows the power consumption profile of the auxiliary equipment of the analyzed rolling machine, which consists of a two roller stand, a gear box, a speed-regulated direct current motor and a hydraulic oil supply for the roller bearings. This profile is helpful e.g. for analysing and optimizing the energy consumption of auxiliary equipment, but also for describing the shutdown, stand-by and run-up behaviour of a process into the ready-to-operate state.
In figure 3 the 134kW motor is ready-to-operate (after about 28s) without the rollers rotating (speed zero).
The consumption while in operation depends on the process parameters and the changing load conditions. Here either measured, static power profiles or parameterized equations can be employed for their description. Parameterized equations provide more flexibility for a broader range of application of the process model to be derived. Figure 4 shows the electric power profile (upper red line) for a load condition of the roller machine at 8 rpm. The lower blue line represents the mechanical power calculated from the roller momentum during hot-rolling. The difference between these two lines further allows to determine the mechanical loss in the idle running power train system and the load-dependent electrical/mechanical process efficiency during operation.
Parameterizing the process models (cf. figure 1) with empirically determined data allows to simulate the power consumption profile of e.g. the hot-metal rolling process as shown in figure 5. In this case a simplified description by constant functions over time was chosen. The power consumption by all auxiliary equipment was comprised to the bottom block in figure 5, which uses energy from start to end assuming a ready- to-operate machine. When the roller speed is raised to the required level, a discrete event adds a rather small amount of power for the now idle running rollers.
The point of contact between rollers and workpiece represents the next discrete event, which adds the load- dependent power of the forming process. Here the simplification of the measured profile (figure 4) to a constant power consumption (figure 5) was applied by keeping the total energy consumption unchanged. For the calculation of the load-dependent power in the hot-forming processes, all necessary parameters as well as material properties, such as parameterized flow stresses, are provided to or calculated in the process models. In this way all entities of the analyzed system (processes, machines and manufacturing chains) with the necessary process control information for the discrete events (from machine control) are described. When the workpiece entity with its geometry description enters a process model (cf. figure 1), all discrete events, according to the loaded NC code, will subsequently be initiated and yield specific energy and resources consumption profiles.
Manufacturing chains analysis and process optimization
In the following, first the energy consumptions of the two manufacturing process chains (cf. figure 2) will be analyzed. Afterwards the results of a parallel manufacturing chain selection and parameter optimization analysis regarding the highest energy and resource efficiency will be discussed.
Figure 10 shows the system entity structure SES of all involved processes and the two analyzed manufacturing chains. From the root of the structure the input (.in) and output (.out) relations between the entities at the next (here top) level is defined in RootDec. This top level comprises three entities, a part source entity, which generates the parts in sequential manner with all necessary parameters (like e.g. material and initial geometry), the process variation entity PV, which presents a link to a sub-structure with the two process chains, and a sink entity, which destroys the part entities at the end of processing, respectively the manufacturing chain. The next level contains the manufacturing chain variants PV1 and PV2. PV1 is for the variant 1 in figure 2 and contains the process entity heating HT, followed by hot-rolling HR, external turning ET (after cooling down) and a combined process description for induction heating and forging IHF. The second manufacturing chain variant PV2 is only made of the external turning process ET. The cutting-off operation (ref. to fig. 2) for separating the workpieces from the bar stock material was ignored in this analysis. The process entities provide data sets for all process parameters which can be analyzed. Using the SES with a specified set of parameters, the discrete events models (cf. fig.1) can be used for calculating energy and resource consumption of the selected manufacturing chain.
Energy consumption of the manufacturing chains
Figure 6 shows the power demand over time, i.e. energy consumption, for all processes of the hot-forming related manufacturing chain (left route in figure 2). The simulation considered a batch of 80 countershafts for which two steel bars of 42CrMo4 (1.7225) with a length of 6 m each were required. The heating of the bars was calculated for a gas powered continuous type furnace with an assumed efficiency of 43% of the employed gas volume. While the second bar was heated, the first bar was already hot-rolled. The considered die forging with local induction heating requires a pre-machining by turning of the two bars with a subsequent cutting off of 80 sufficiently long billets.
The last power profile in figure 6 shows the partial induction heatings (9 min., 1100°C) followed by the die forging peaks for the shaping of the shaft shoulders of each of the 80 countershafts. The total manufacturing time for the highest resource efficiency of this chain was calculated to 764 minutes.
The second manufacturing variant considered only the turning process after cutting-off 80 bar sections. Figure 7 shows the power profile of the first 10 countershafts. The shaft design requires two clampings for turning both sides of the shaft and the same number of cutting passes. Between turning the shafts a pause of 1 second was included. For a constant cutting velocity, the speed was raised stepwise with the reduction of the shaft diameter, which led to the slight power increase during the sequence of cutting passes for each shaft side. The total manufacturing time for the highest resource efficiency of this chain was calculated to 392 minutes.
Manufacturing chain selection and optimization by fitness function and genetic algorithm
For the analysis of the most energy and resource efficient manufacturing strategy in combination with a parameter optimization, the system entity structures of both manufacturing process chains were analyzed together. The chosen optimization parameters were the manufacturing variant, represented by the manufacturing chains, the roller speed, the hot-rolling respectively furnace temperature, the cutting velocity, depth of cut and feed rate of the turning processes. The resulting energy and resources consumptions for all processes and parameter combinations were calculated and weighted by their specific cost factors, i.e. prices per unit, and summed up in a fitness function (eq.1).
G = A*Welect + B*Wgas + C*mwork + D*VCL + E*Ntool (1)
The analyzed resources were the necessary electrical energy Welect , the amount of furnace gas energy Wgas , the removed volume respectively mass of workpiece material by cutting mwork , the loss of cutting lubricant VCL and the number of cutting tools Ntool. Hence the value of the fitness function represents the energy and resources costs of the whole batch for the analyzed part of the manufacture. Dividing the cost contributions by the batch size, the price per shaft is calculated.
For an effective determination of the best combination of process parameters and manufacturing chain, the Matlab genetic algorithm (GA) was employed towards minimizing the fitness function. The options of the genetic algorithm are listed in table 2.
Figure 8 shows the course of the varied process parameters, the resulting resources cost contributions per shaft and the total costs per unit shaft until the tolerance criterion of the fitness function G was met.
The upper half of figure 8 shows the course of the varied process parameters over a total number of evaluated simulation experiments of 250. It can be seen from the upper left diagram, that the genetic algorithm (GA) switched several times between evaluating manufacturing chain 1 (with hot- metal forming) and chain 2 (only cutting) until it remained only in the first one.
While the furnace and hot-rolling temperature as well as the roller speed were varied only for manufacturing chain 1, the cutting parameter vc, ap and f were varied in both chains until the end, the feed rate f in particular.
The lower part of figure 8 shows the resulting resources costs and total costs per unit shaft over the course of simulation experiment evaluation. As gas furnace and induction heating are only used for hot-metal forming, the costs for electricity and gas are up for this manufacturing variant. The main costs for the turning variant (2) result fromthe material removal (swarf) and the tool costs (cutting inserts, cf. peaks for chain 2 around experiment no. 150). Figure 9 summarizes the optimization results in terms of costs per shaft for the analyzed processes, showing also the potential of the presented approach relative to reasonable initial start parameters.
Using a fitness function and a genetic algorithm, an effective selection and optimization of manufacturing chains was realized for processes described by discrete event models and their relations by system entity structures.
The resources efficiency oriented selection and parameter optimization for two model manufacturing chains has let to favouring the hot-metal forming variant over the cutting variant, although the energy efficiency of the cutting variant is much better within the analyzed system limits. The larger material removal resulted in an expensive work material loss (waste) and higher tool costs for the cutting variant. The optimization reduced the cutting lubricant and energy costs by shortening the cutting process because of a larger depth of cut and feed rate.
The highest energy saving potential in hot-metal forming has the reduction of the furnace and rolling temperature, respectively the forging temperature. However it needs to be verified if the higher process forces, due to the lower forming temperatures, are acceptable regarding tool and machine load as well as tool wear and workpiece quality.