Forecasting and Planning for Material Control in the Medical Device Industry

,


Introduction
PT XYZ is a domestic company engaged in manufacturing medical devices located in Osowilangun, Surabaya.Products produced by PT.XYZ has approximately 251 products with 36 categories.Apart from having many types of products, PT.XYZ has varied and fluctuating product demand.This request means that simple planning is only enough to fulfill part of the order.The inability to carry out forecasting adds to the problems that make companies overwhelmed in predicting demand, managing production activities, and ensuring stock availability, causing overstocks and stockouts (Silver et al., 2016).Figure 1 shows the sales graph or demand graph for 12 types of product components at PT. XYZ can be seen in the sales graph for the past 1.5 years experiencing uncertainty at all times.To meet customer demands and reduce the occurrence of loss sales PT.XYZ always tries to maintain and adjust its production activities.One factor is managing, maintaining, and determining the quantity and quality of product components stored in the inventory warehouse.The storage value can reach a percentage of up to more than 25% of the total value of the company's assets.Every service or manufacturing company needs storage because it will reduce the risk of customer demand not being met (Pujawan & Mahendrawathi, 2017).
At PT. XYZ's inventory management process is carried out by the Production Planning and Inventory Control (PPIC) section by updating stock for product components.PPIC will look at the warehouse stock report and then match the existing components with the assembly formula (Chopra et al., 2016) .When the availability of products that can be produced is less than the number of purchase orders (PO) received, an order will be placed with the supplier (Kovačević et al., 2020).After comparing stock data and PO data, the shortage data is forwarded to the purchasing department for further processing and obtaining approval from the main director to make import purchases.The lead time for component delivery takes 2 to 3 weeks, after which it will be received by the warehouse and the suitability of the components will be calculated regarding the quantity and quality of the components ordered.After all are declared to Original Article have passed testing by the QC, the components will be stored in the warehouse according to each product type (Sohrabpour et al., 2021).
The number of electromedical products that have been produced bought and sold is approximately 140, but currently, there are approximately 50 products that are still active and frequently ordered.Some products include CMS-600 PLUS, ECG-1200G, TENSIONE, ECG-300G, SON-C, SON-B, SON-PRO, USG PROMAX, PROMIST-3, MED-01, ECG-100G, and ULTRAMIST.These products are ordered to suppliers in the form of assembly components in 'sets' so that the quantity of products can be produced as many as the 'set' ordered.In the last 1.5 years from 2022-2023 June, the demand and assembly graph for these products can be seen in Figure 2. It can be seen that the comparison between demand and product assembly has quite a difference.Many more SON-B products were assembled than the total quantity requested, this shows that the purchase of product components exceeds the estimated needs and makes storage costs high.Apart from overstock problems, stockout problems also occur in several products such as ECG-1200G, TENSIONE, SON-PRO, and PROMIST-3.These products have an assembly quantity that is less than demand, this indicates that there is unfulfilled demand which causes losses and a decrease in service level values even though this calculation has not been applied to PT.XYZ.With the existing storage management policy, PT.XYZ will be in an uncertain and ineffective state because all decisions are based on inaccurate data processing and rely solely on intuition.Large orders for product components are made without accurate forecast calculations, so excessive or repeated purchases often occur which will incur inventory costs.This can happen because PT.XYZ does not yet have a good inventory management system such as safety stock, reoder point, maximum capacity, and several component orders (Tersine, 1994;Taha, 2017).From this existing condition, it is necessary to have inventory management that can minimize the occurrence of overstock, and stockout and minimize storage costs so that the desired service level value is obtained (Bueno et al., 2020).
In this research, forecast calculations will be carried out using the Croston and Syntetos-Boylan Approximation (SBA) methods by the type of demand from the results of product component classification using the Average Demand Interval (ADI) and Coefficient of Variations (CV) methods (Sirisha et al., 2023).Next, calculations are carried out related to inventory policy using the continuous review method (s, S) and (s, Q), the results of which will be compared with the existing inventory policy using Monte Carlo simulation to find out which policy is the most effective and efficient, especially in terms of total cost and service level resulting from.

Methods
This research methodology consists of literature studies, field studies, data collection and data processing, ADI-CV calculations, forecasting calculations, forecasting accuracy calculations, comparison of forecasting results, calculation of total inventory costs, comparison of total inventory costs, Monte Carlo simulation, analysis and interpretation of results, conclusions (Baskhara, 2022).

Classification
Classification can be calculated using the Average Demand Interval (ADI) and Coefficient of Variations (CV) value parameters.
With: N = number of periods without the value 0 ti = the interval between 2 consecutive non-0 demand periods

Forecasting and Errors
The Croston and Syntetos-Boylan (SBA) approach is very useful in forecasting demand for goods with irregular time intervals, such as demand for rare goods or goods with sporadic demand patterns.The Croston method calculation consists of two main stages: a. Calculation of Actual Demand Levels With:   = estimated actual demand at the time t   = actual demand at the time t  −1 = estimated actual demand at time t-1  = smoothing constant that can be selected according to needs (usually between 0.1 to 0.3) b.Calculation of Time Intervals Between Requests With:   = the time interval between actual requests at time t  = the time when the actual request occurs  −1 = the time interval between actual requests at time t-1  = smoothing constant that can be selected according to needs (usually between 0.1 to 0.3) Meanwhile, for the SBA, it is as follows: In this system, the quantity (Q) ordered is a fixed quantity when inventory falls to reorder points or lower.The following are the steps for calculating the system (s, Q): Step 1 Step 2 Step 3 Step 5 Step 6 Replenishment is also carried out when inventory falls to the reorder point or lower.However, unlike the (s, Q) system, the replenishment quantity in this system varies and is not fixed to increase the quantity to point S. The following are the calculation steps for the (s, S) system: Step 1 Step 2 Step 3 With: Q = order quantity SS = safety stock

Monte Carlo Simulation
In this simulation, it is necessary to determine the probability distribution of each variable to be studied.The assumptions used are that demand is normally distributed and lead time is constant.Meanwhile, in PT XYZ conditions, the distribution of demand data and lead time is not constant.So this Monte Carlo simulation will be used as a test of the calculations carried out in the previous subchapter, where the simulation will adjust to fluctuating demand and non-constant procurement lead times (Ghobbaret and Friend., 2002).

Results
Calculating the percentage of total storage costs to determine the storage costs for product components per unit of time is obtained from the sum of storage costs, insurance costs, capital costs, and loss costs.Next, the percentage results are multiplied by the product component costs to obtain storage costs as in Table 1 and Table 2  Next, determine the product classification based on historical demand data using ADI-CV to determine the type of demand so that you can determine a suitable method for forecasting calculations at the next stage as in Table 3 and Table 4 Croston and SBA From Table 3 and Table 4 it can be concluded that all types of product categories are lumpy and erratic demand, which means the forecasting methods used are Croston and SBA as in Table 5 is an example of forecasting calculations for CMS-600 PLUS product components then the final determination of method selection is seen from the error value by comparing MSE and MAD as in Table 6.USG PROMAX 0,3 0,9 0,9 0,6 0,8 0,5 0,9 0,9 0,5 0,7 0,7 0,9 0,9 0,4 0,6 0,9 0,9 0,9 0,3 0,5 In Table 6, forecasting accuracy tests are carried out using MSE and MAD so that the forecast method is determined using the smallest error value.For CMS-600+, ECG-300G, USG PROMAX, and MED-01 products use the SBA method and the rest use the Croston method.After that, policy parameters were calculated and policy results were compared using Monte Carlo simulation to get the best results based on total costs and service levels as in Table 7 -Table 12.     12 are the results of the Monte Carlo simulation for all inventory management policies.The results obtained are that the Continuous Review s, Q policy produces the lowest total cost of Rp. 69,881,238,792 with the highest service level of 99.26%.Meanwhile, other policies have higher total costs and lower service levels.The existing policy is caused by higher holding costs and stockout costs due to storing too many product components in the warehouse, but on the other hand, it cannot meet demand because supplies have run out for a certain time.Meanwhile, the Continuous Review s,S policy is caused by order costs and unit costs that are too high due to the process of ordering quite a lot of products.
Retesting was carried out to determine the policy's resilience to uncertainty by providing a margin of error of ±20%.The policy chosen is Continuous Review s, Q with a comparison of the results of existing conditions and improvement policies in Table 13 and Table 14.Table 13 and Table 14 show the positive results of policy improvements from implementing selected policies Continuous Review s, Q.The selected policy can increase the service level value by 1.36% and reduce total costs by IDR 22,206,322,981 compared to the existing policy.

Conclusion
The data collection experiment was carried out using 4 α values (0.3, 0.5, 0.7, and 0.9) which shows that the greater the α value, the smaller the error produced.The error calculation uses the MSE and MAD methods so that 4 product components (CMS-600 PLUS, ECG-300G, USG-PROMAX and MED-01) are obtained using the SBA method and 8 product components (ECG-1200G, TENSIONE, SON-C, SON-B, SON-

)
Measurementof forecast error a. Mean Squared Error (MSE)   = |  |  = data error  = lots of data b.Mean Absolute Deviation (MAD)   = |  |  = absolute error   = data error  = lots of data Inventory Management a. Continuous Review (s,Q) 22) With:  0 = order quantity  1 = order quantity taking into account N(k)  = number of requests () = demand probability  = backorder costs  = message cost ℎ = holding costs  = safety factor () = the size of the inventory shortage   = standard deviation of demand during lead time E(k) = partial expectation  = 0,1  = reoder point  = standard deviation of demand  = mean demand  = lead time b.Continuous Review (s,S)

Table 1 .
. Percentage of total storage costs

Table 2 .
Storage costs/product components

Table 4 .
Forecast methods based on component classification categories

Table 5 .
Calculation of CMS-600 PLUS component forecasting using the Croston and SBA methods

Table 1 .
Forecasting Accuracy Test with the Croston and SBA Methods

Table 7 .
Existing policy parameters

Table 8 .
Recapitulation of continuous review parameter calculations (s, S)

Table 9 .
Recapitulation of continuous review parameter calculations (s, Q)

Table 10 .
Comparison of simulation results of total costs from existing policies and selected policies

Table 11 .
Recapitulation of simulation results for improving continuous review inventory management policies s, S

Table 12 .
Recapitulation of simulation results for improving continuous review inventory management policies s, Q

Table 13 .
Comparison of total cost simulation results from existing policies and selected policies

Table 14 .
Comparison of service level simulation results from existing policies and selected policies