Operating cost estimation and analysis for polymer pro-

2.3 Application examples

2.3.4 Operating cost estimation and analysis for polymer pro-

Having cost information from e.g. the conceptual design phase of a technology used for economic evaluation, one can simulate the operating costs and detect relationship between several operating variables. For this purpose a simple cost model needs to be evaluated. Here, not only SOM is an applicable data mining solution but other exploratory data analysis tools are at hand to monitor the process (see Section 1.2.2). As a simple energy-cost model, two main components of production cost were identied with one product revenue value:

1. Energy costs:

- monomer feed pumps (assumed to be constant during a production) and circulating pump energy costs,c_mp=c_cp = 20 HUF/kWh;

- cooling jacket water cost, cjw = 8 HUF/t;

2. Raw Material costs:

- Propylene, cC3 = 220 HUF/kg;

- Catalyst, c_CAT = 80 000 HUF/kg;

- TEAL, cT EAL =1 500 HUF/kg;

- Donor, c_Don =2 500 HUF/kg;

- Hydrogen,cH2 = 270 HUF/kg;

3. Product revenue: Polypropylene,cP P = 300 HUF/kg;

The following equation was set in order to estimate production cost:

C = (FC3,R201+FC3,R202)cmp

+(Qpump,R201+Qpump,R202)ccp

+(F_jw,R201+F_jw,R202)c_jw +P

iF_mat,ic_mat,i

(2.5)

Note that in the above equation, the monomer ow was considered to be equal to polymer production regarding a utilization factor of 1.02 (appr. 2%

monomer loss in the process), and in the energy-cost model the natural gas

and steam consumptions are neglected because of lack of data available on DCS (thus it could not be collected into DW).

Within a production period in 2006, altogether 12 homopolymer produc-tions were selected for further investigation. Four dierent polymer grades were produced: H1 twice (H1a and H1b), H3 twice (H3a and H3b), H4 ve times(H4a to H4e respectively) and H5 three times(H5a, H5b, H5c).

Box-plots were applied to visualize the specic production costs of a given type of product (H4 in this case) presented in Fig. 2.22 and 2.23. As these plots show, for the same type of production the costs are almost constant and the production is stable: small box size and small min-max ranges can be detected. However, one can detect that for H4c production, the specic material cost is lower, which indicates that there should be an error in the data: the data analysis has shown that during that production the catalyst ow measurement failed (its cost addition was almost zero).

Energy cost for 1 kg PP (Ft/kg)

H4a

Figure 2.22: Specic energy cost for H4 homopolymer productions.

From the analysis it can be seen that the production cost depends far the most on raw materials consumption - as expected. Energy cost takes a little part from the costs, but it gives more freedom to the operators and thus it can be a place for improvement. In Eq 2.5, cooling water supply is responsible for appr. 60% of energy related cost while 35% goes for circulating pumps of the reactors. In this section we focus on these two elements to be analyzed.

For dierent types of homopolymer grades, the dierent productions need dierent amounts of energy as Figure 2.24 shows. It is not because of the MI dierences (MI stepwise lowers from H1 to H5) but it rather has operational issues - a point where operational improvement can take place.

Cost of producing 1kg PP (material)

H4a

Figure 2.23: Specic material cost for H4 homopolymer productions (this vi-sualization has detected a measurement failure in catalyst ow in third case).

Energy cost for 1 kg PP (Ft/kg)

H1a

Figure 2.24: Specic energy cost for dierent homopolymer productions.

To detect process variables that are responsible for energy cost, quantile-quantile plots are ideal tools. A priori knowledge can be incorporated as well because it is known that circulating pump power depends on slurry density, which can be manipulated by monomer feed to the reactors (see structure of lo-cal control on Fig 2.5). This energy cost-monomer feed correlation can be seen on Figure 2.25 for the rst loop reactor. Unfortunately, power consumption of propylene feed pumps are not collected on DCS. Their power consumption were modeled via AspenTech's AspenPlus steady simulation software to con-tribute to the energy cost calculation. However, they were neglected in this study having steady values for a production. It will not cause too much

er-ror because of only the 3-5% of the energy costs, which is acceptable in this