Adaptive lean and green approach for continuous improvement

The static analytical method has been widely illustrated in many case study associated with output prediction. However, the static analytic method is unable to cope with the dynamic changes in the real world. The need for an adaptive analytical model with an updating algorithm with time step can be beneficial in addressing the dynamic challenges. The constant feed-in of data from the organisation can be beneficial to the model in predicting a better outcome. The continuous updating algorithm is also known as the adaptive feature that responds over time. The back-propagation (BP) algorithm, which operates based on the reverse mode of differentiation (Griewank, 2012), is incorporated in the adaptive model. Hie BP method is then popularized with an application on neural networks by Rumelhart et al. (1986).

In the analytical lean and green (L&G) approach, the expectations for the lean and green index (LGI) can change with tune. The accessed LGI is constantly being collected with respect to the expected index. The update rules are being utilised by the BP algorithm to compare the difference between the accessed

LGI and expected LGI for every time step. The error used is the mean squared error, which is defined as the following:

where E is the error and LGI can be an accessed or expected index. As an improvement strategy is planned, the expectation index can be targeted by the industiy player prior to implementation. On the other hand, the processing data are evaluated in order to obtain the accessed LGI after the improvement strategy has been executed. The gradient descent method with chain rule is applied with the use of BP. Equation (36) indicates that the updating of AHP weight for the intermediate layers (4M1E) is computed:

1V represents the weight of the five main components (4M1E) on the current month. w_w is the new weight (target) for the next month, determined by the organisation. u’_H represents the weight for the weight of the previous month, while E reflects the error having subscripts of the same notation. A proportional constant, which known as the learning rate //, is used to update the weights of each components. Wilson and Martinez (2001) discovered that the learning rate of 0.05 reflects the smallest error, indicating that this is the fastest learning rate in terms of reduction in total sum square (tss) error.

k₎ are the weights for the each of the indicators under the 5 main components, where ; is the time step for current month. The calculated error is passed down to the analytical model with the chain rule and updated for each time step.

The weight update for the components and sub-components are critical in performing performance iteration. The continuous iteration on the component’s weight will reflect on the areas that can be further improved by the organisation. Thus, the updating process is important for the organisation and the analytical model to achieve the target LGI.

Lean and green case study

An application of the lean and green (L&G) model is demonstrated with a cogeneration (COGEN) case study.

Figure 5. Flow diagram of case study.

In year 2015, the COGEN plant started its operation independently without synchronising with the national grid. The requirements of the COGEN plant are shown below:

a. Main fuel source : Natural Gas

b. Maximum electricity generation : 6.5MW @ 32degC

c. Maximum steam recovery : 16 tonnes/hour @ lObarG saturated steam

d. Total operation days : 351 days

e. Total operation hours : 8424 horns

f. Annual shutdown days : 14 days

Data collection and discussion

Figure 6 demonstrates the batteiy limit of the COGEN case study. The 5 main components (i.e., manpower, machine, material, money and environment) are reflected in Figure 6. The data collection is executed as per Figure 2. In this study, operational data of the plant from the year 2015 to the year 2018 are collected. According to the operation manager, some of the indicators (i.e., rate of recyclable defects and inventory management) in level three (in Figure 2) are not applied in this study, as reflected in Figure 2.

Figure 6. Boundary limit of COGEN case study (Leong et al., 2019a).

Analytic hierarchy process (AHP) analysis

Table 2 demonstrates the distribution of the 5 main components’ (4M1E) weights. Based on the industry expert’s input, money (MY) was ranked first in improving LGI of COGEN plant (26.10%), followed by environment (EV) (21.54%), manpower (MP) (18.7%), machine (MC) (18.28%) and material (MT) (15.38%). The industry expert emphasizes the importance of experienced personnel in managing the COGEN facility due to the lack of experienced talent in the COGEN industry. This has increased the challenges for an organisation in retaining the talent.

Table 3 illustrates the priorities of manpower (MP) indicators. According to the industry expert, the compliance of occupational health and safety of the employee are the utmost priority of an organisation. The operation manager added that continuous improvement training can improve the level of competency of the employee in order to allow the employee to perform their duty safely'.

Table 2. Consolidated pairwise comparison matrix of 4M1E with respect to the goal.

CR: 0.0283.

Table 3. Pairwise comparison matrix of manpower (MP) indicators.

CR = 0 063.

Table 4. Pairwise comparison matrix for money (MY) indicators.

On top of that. Table 4 shows the dominance factor of money (MY) where profit indicators are deemed to be more significant in comparison to operation cost. Based on the operating team, the maintenance of the system has been contracted to a qualified contractor at a monthly fixed cost.

Lean and green index (LGI)

Figure 7 reflects the historical trend of manpower (MP) performance. The lower competency and key performance index (KPI) score are mainly due to the initiation of the new process where the MP has minimal experience in handling the COGEN process. To ensure the COGEN team can manage the operation effectively, extensive training and guidance are provided, ensuring the competency of the employee. In Figure 7, it can be observed that the improvement in competency rate has also improved the rate of achievable KPI of the employee. This reflects the overall improvement in MP performance.

Figure 8 reflects the indicator for the machine (MC). It is observed that the MC index has been improved substantially in the year 2017, as the COGEN output demand has been maximised. As informed by the operation team, the COGEN facility operated at lower demand rate in the years 2015 and 2016 due to lack of energy demand from the end-user. In the year 2017, the MC index also achieved an overall equipment effectiveness (OEE) of higher than 85%, which is considered as a world-class performance.

The input resource and output resources performance of the COGEN plant are reflected in Figure 9. The MT index tends to be lower in the years 2015 and 2016 because the demand of the COGEN facility was low. On top of that, the COGEN facility requires a minimum fuel input to maintain the idle condition. In this case, the storage and inventory requirement for the COGEN facility is not required because the fuel source (i.e., natural gas) is supplied directly from the gas pipeline.

Based on the expert’s input, money (MY) index is the top priority in the COGEN facility. Referring to Figure 10, the profit rating of the year 2015 is relatively low compared to other years. Based on the operation manager's explanation, the performance in the year 2015 is mainly due to a lack of experience in operating the plant.

Lastly, Figure 11 demonstrates the performance indicator of the environment (EY). The scoring for carbon footprint indicator has improved. The overall EY index has increased as the performance of the COGEN system has improved. The EV performance is highly dependent on the MC performance, as the dry low emission (DLE) mode depends highly on the load demand of the prime mover in order to maintain the low NOx emission.

Figure 12 and Table 5 show the L&G outcome of the case study. Generally, a consistent improvement in LGI can be observed from 2015 onwards. The carbon performance reflected from EY takes tune to reflect after improvements to the other components have been made. Therefore, the improvement on 4M1E

Figure 7. Manpower (MP) indicators.

Figure 8. Machine (MC) indicators.

Figure 9. Material (MT) performance indicator.

shall be monitored progressively in order to ensure that the L&G approach is effectively implemented. In this case study, carbon management is challenged by the competency of human resource. Figures 7, 8, 9, 10 and 11 have shown that the operational and environmental performance improved with better employee competency rate. Thus, the carbon management strategy is highly dependent on MP planning.

Figure 10. Money (MY) performance indicator.

Figure 11. Environment (EY) performance indicator.

Figure 12. Progress of L&G performance index (Leong et al., 2019a).

Table 5. Summary of L&G index.

Analytical continuous improvement in the lean and green index

According to the operation team, the COGEN facility has been operating under the same operation strategy since 2015. In July 2018, the backpropagation (BP) optimisation, where the COGEN facility is required to established a monthly target, was introduced. Figure 13 reflects the LGI performance after the implementation of BP optimiser. The BP optimiser evaluates the feedback on improvement done based on the five main components and its sub-indicators. The performance of new changes or action done in the operation will be reflected in the LGI in Figure 13. It can be observed that, starting from August 2018, the monthly LGI indicates an oscillation pattern as the BP optimiser provides feedback to the industrialist for improvement.

Figures 14, 15 and 16 show the weight update of 4M1E components, MP indicators and MY indicators based on BP optimisation, respectively. The figures are indicating a similar graph pattern as the BP optimise the L&G model. Referring to Figures 14, 15 and 16, the BP optimiser explores the potential improvement area by using the previous data points in the months of September and October 2018. With the continuous feedback from BP optimiser and operation team, the BP optimiser has shown signs of data convergence in the months of November and December. The convergence of the indicators is evaluated by the gradient descent method, where the optimum point of the system is identified. The outcome of the converged values is similar to the initial weight, this indicates that the experts have provided useful insights on the operation. The generated results from BP optimiser are entirely reliant on the input of the process parameter data. Thus, BP optimiser can enhance an organisation’s performance through continuous process improvement.

Figure 13. Monthly LGI update (BP optimisation initiated in July).

Figure 14. Backpropagation optunisation on 4M1E mam components.

Figure 15. Backpropagation optunisation on MP indicator.

Figure 16. Backpropagation optimisation on MY mdicator.

6. Conclusions The efficiency and effectiveness of traditional manufacturing practices are no longer competitive in this global trading world. Adding on to global warming and climate change, the enforcement of stringent regulations and policies har e challenged the industrialists to seek out alternative technology in order to improve their environmental performance. With the implementation of the adaptive lean and green approach, the industrialists will be assisted in the reduction of non-value-added product from the operation and environmental based on 4M1E. This will not only enhance global competitiveness, but also improve the carbon management system of the organisation. The adaptive method allows the organisation to practice continuous improvement and track the improvement rate by constantly exploring potential improvement areas. The integration of L&G approach not only reflects a positive outcome in carbon management, but also performance improvement. The adaptive approach will be able to predict and provide a better performance improvement indication with the increase in the operation database.