Swarm Intelligence For Resource Scheduling: A Case Of Manufacturing Industry Essay.

Write a report on the Swarm Intelligence for Resource Scheduling: A Case of Manufacturing Industry.

Machine intelligence or Artificial intelligence has emerged as one of the major sectors that is making contribution in simplifying challenges faced by the humans in different sectors. The sectors in discussion includes a broad range of industries or more appropriately every industry that is dependent on machineries for the completion of their operations. The operations include the organisational processes, different management processes and other processes as well (Chen, Chen & Chang, 2013). One of the most prominent areas that is being greatly supported by the machine learning is the Project Management. However, operations management within an industry is a very vast domain and in almost every aspect of the domain the machine learning is offering its support. One of such aspect is the resource scheduling which is one of the most crucial aspects and have been selected as the area of research.

  Machine learning has also evolved with time and it is now considered to be an umbrella which have different algorithms with improved capabilities and features are making themselves the solution in context with the automated process. One of such algorithm that has established itself is Swarm Algorithm (Faria et al., 2013). Swarm Intelligence (SI) is integral part of field of artificial intelligence which is used for higher complexity problems. SI is used to solve non-linear design problems with a real world application which makes it crucial for different applications (Pakirao & Bose, 2018).

Hence, the discussed paper has detailed the role of the subject (Swarm Intelligence) in resource allocation and job scheduling within a manufacturing industry. The paper has been initiated with introduction where the background of the paper has been established. The next section has detailed the problem that the paper has attempted to answer which is succeeded by the potential solution to the problem statement and the approaches that can be adopted to mitigate the threat. Finally, the paper has been summarised to conclude on the discussion.

Different aspects are crucial pre-initiation of an activity or operation that are aimed at ensuring that the operation being undertaken is efficient and in the process a high productivity is gained from the process. One of the key factors is the resource scheduling which can be categorised in two categories namely, job scheduling and resource allocation (Zhan et al., 2015). There are multiple challenges associated with both of the aspects being discussed. Job scheduling needs to take account of factors such as the schedule preference, skills required, qualification and training, availability and others. Furthermore, the labour laws that are being enforced by the government around the world for prevention of labour rights violation also needs to be taken in account before scheduling the job (Du et al., 2018). Considering all of the factors discussed above before scheduling the job makes it one of the most complex task that needs to be mitigated.

Additionally, resource allocation to a project or activity is also one of the most complex task because the firm needs to ensure that the resource allocated by them to an activity is being utilised and maximum output is derived from it (Gahm et al., 2016). The need of resources, availability of resources, inventory capacity, resource management and different aspects along with the activity needs are taken into consideration before allocating resource which makes it one of the most complex task in successful completion or efficient operation of a process. Hence, the problem statement of the paper is to mitigate the complexity of resource allocation and job scheduling with assistance of technological advancement.

Potential Solution

Several models, frameworks and technological advancements are available to address the discussed problem statement. Adequate operation management, automated resource & job scheduling, use of software for the discussed purpose and several others. One of the potential solution that had catch eye of the manufacturing units around the world is the process of automation through machine learning that is AI (Artificial intelligence) (Russel & Norvig, 2016). AI, based on the strategies adopted, collected data and the actions taken by the human factors in the past, the machine starts learning and accordingly plans and schedules a project. With time the machine understands the concepts and based on the collected data makes the plans and schedules. AI has is also a very broad terminology and one of the integral part of AI that can offer great relevance to the area in discussion is SI.

Swarm Intelligence (SI) was first introduced in the year 1989 by Jing Wang and Gerado Beni (Zhang et al., 2014). It is an integral part of field of artificial intelligence which is used for higher complexity problems. SI is used to solve non-linear design problems with a real world application. SI basically refers to the collective behaviour of self-organised and decentralised systems. It can be natural or artificial in nature and is based on the concept of artificial intelligence. The operations of the SI are done through the agents or boids that are part of it and are responsible for the communication at a local level and even with its environment (Sung et al., 2017). The discussed intelligence approach is inspired from the biological systems where the ants, birds and others shows a coordinated action based on the response from the previous being. The agents abide by simple ethics and regulations and as there is no centralised system to depict the agents in the actions they take. Their responses decide their behaviour which is random along with the interaction among them, an “intelligent” global behaviour is emerged.

The discussed intelligent technology has shown its efficiency in different genres that includes the forecasting of problems based on the local and environmental interactions. It is commonly termed as “Swarm prediction” (Kumaravel & Sengaliappan, 2016). Other applications of the discussed technology are evident in the robotics industry, crowd simulation and other prominent sectors. Hence, based on the predicting nature of the technology along with its crowd simulation capabilities, it is stated that the technology will be suitable for resource scheduling. The following section is dedicated to the discussion over the role that the subject (swarm intelligence) is capable of playing in the resource scheduling of manufacturing industry.

Swarm intelligence optimisation should be understood to get an insight into the role that the former is capable of playing in making the most effective and efficient use of the presented scenario or resource. Hence, the following subsection has presented an insight into the optimisation methods presented by the swarm intelligence.

Ant Colony Optimisation:

As the name of the optimisation methods suggests that the discussed method is inspired by the ant colony and their natural behaviour (Mirjalili, 2019). Hence, to understand the method three crucial points needs to be understood and that includes

1. Ants pursue pheromones to determine the shortest path between the food source to their home and vice versa.
2. It should also be noted that pheromones evaporate quickly.

• The ants prefer shorter path that have the highest concentration of pheromones of all the available paths.

Approaches

The discussed optimisation technique aims at identification of optimal path through the graph based on the pattern that ants equip to identify the shortest path. The next step reflects the evaporation variable (ρ) and how much percent pheromone evaporates in every iteration (Golding et al., 2017). It also reflects on the pheromone left behind by each ant on their respective trail through the variable Q. The final variable is that after each iteration the pheromone value is updated by m number of ants who were part of the solution determination. Here the iteration of solution is reflected towards determination of the shortest path.

The optimisation techniques in discussion is used in different aspects of industries where the agents replace the ants and the pheromone is replaced by the interaction between the agents and the environment. As part of the whole process the agents keep track of all the specific solutions that they have identified and left a trail behind. In the discussed scenario, the agents or boids accumulate a virtual trail on the path segments. The paths are selected at a random basis which is in accordance with the trail presented on the possible path from the initial node. The agent reaches the following node and adopts a new path and continues till it reaches the initial node. The completed tour is the solution and the tour is analysed for optimality.

Figure: Construct Ant Solutions

(Source: Golding et al., 2017)

The image attached above reflects the solution that has been determined by the ants during their course where the ant moves from node “i” to “j”. τ_ij represents the amount of the pheromone, α is the influence control parameter of τ_ij, η_ij is edge desirability, β is the influence control parameter for η_ij.

While moving along the path the ants also update the pheromone amount that is governed by the equation shown in the image below

Figure: Construct Ant Solutions

(Source: Golding et al., 2017)

τ_ij is the pheromone amount on the nodes, ρ being the pheromone evaporation and ?τ_ij is the amount of pheromone that has been deposited.

The most prominent benefit of the discussed method is that it offers inherent parallelism and the positive feedbacks assists in rapid discovery of suitable solutions.

Particle Swarm optimisation inspired by the flocking of birds or schooling of the fish was developed by Ebenrhart and Kennedy (Jordehi & Jasni, 2015). The first step that is part of the discussed optimisation process is the development of the initial particles and are assigned with some initial velocity. The next step as part of the optimisation process includes evaluation of the objective function at location of each particle. The discussed step is undertaken to determine the most suitable (lowest) function value and accordingly best location. Furthermore, it assists in selection of new velocities based on the velocity of the particles at that instant moment, best location of the individual’s particle and most proper location of their neighbours. The next step involves iteratively updating of the particle locations (that is the new location will be the old in addition to the velocity along with modification to ensure that the particles stays within the boundary), neighbours and velocities (Sun et al., 2016). The iteration will be a continuous process until the algorithm will reach the stopping criteria.  

The algorithm is governed by the equations that is evident from the image attached below.

Figure: Particle Swarm Optimisation

(Source: Sun et al., 2016))

Here, c₁ & c₂ are the positive constant and a random function within the domain [0,1] is presented by rand(). X_i is the reflection of i^th particle, P_i is the previous position, V_i is the rate of position change for the i^th particle (Sun et al., 2016).

Bees algorithm is a search algorithm based on the population that was first introduced in 2005. The algorithm is inspired by the food foraging behaviour of the honey bees and based on that behaviour the decision making is enabled in different computational problems (soufyane Benyoucef et al., 2015). The main idea of the optimisation method is to determine the best similarly as the bees determine the best site for nest from many sites by taking in account accuracy and speed. The decision for the site and food is not dependent on some individual bees but a group decision is made and similarly, the optimisation methods considers different findings into account before concluding on a decision. The bees communicate through on waggle dance and in the process shares precise information of distance, direction and quality of the designation (Ozturk, Hancer & Karaboga, 2015). Based on the same the discussed algorithm operates and initiates with initialising the population with random solutions and evaluates the population fitness. In the process another measure is adopted and that is if the stopping criterion is not met the forming of new population is a continuous process. The next measure involves selection of the sites for the neighbourhood search and recruiting of the bees by evaluating their fitness in the process. The fittest bee from every batch is selected and the remaining are assigned with the search duty where they randomly conduct the search and evaluates the finesses as well.

In the discussed scenario, the bees are referred to the agents which are designated with identifying the solution and the most suitable solutions ae adopted while the remaining agents continue their search for the best possible outcome. The image attached below reflects the basic flowchart of the bee based algorithm.

(Source: Ning, Zhang & Zhang, 2016)

Scheduling is a process of decision making that is crucial for different industries and manufacturing industry is one of them. Furthermore, from the discussion above it is evident that the swarm algorithm is one of the best method for decision making and hence, the sub-sections below has detailed the role that different optimisation method is capable of playing in resource scheduling of the manufacturing industry.

  In the discussed approach the node pheromone is used for the calculation of the probability of the path choice. The process accounts for the amount of the resource exhausted by the agents (artificial ants) throughout the path the amount of pheromone is reduced and in the process enabling the agent community to explore different paths (Mirjalili, 2019). The opportunity of exploring different solutions along with the probability of optimal solution and achievement of the load balanced system is done through the measure discussed above. After the completion of a task of the designated resource node, the pheromone is updated by the algorithm over the path that is shortest and most efficient (Rohini & Natarajan, 2016). It enables the real-time preparation of next task through the use of update mechanism that is dependent over the success ratio of the previous task.

It is evident from the discussion above that the scheduling of the task is real-time and dependent on the previous task which could be taken into account to state that the discussed approach is adaptive in nature and the benefits along with challenges offered by adaptive resource scheduling is applicable on the discussed approach as well (Campos Ciro et al., 2016). Hence, the allocation of the resource can be changed depending upon the progress of the task. If the task is failing additional resource could be added, reduced or changed accordingly. The discussed approach is also capable of determining the reliability and efficiency of the predecessor task. It can be attained by measuring the trustworthiness of the resource node. It is measured by the incentive or punitive factor. If the task returns to its initial position successfully, then the task is considered to have incentive factor while in the contradictory scenario the task is considered to have punitive factor.

The discussed method also accounts for the failure scenario where, if a task is failing than the rescheduling mechanism is activated. Rescheduling mechanism is the mechanism that is part of the algorithm that reschedules the tasks depending on the success level of the continuous task (Kalinowski et al., 2017). The discussed measure is aimed more at protecting other nodes from getting infected by the failed task rather than protecting the task that is moving towards failure. It can be considered that through the discussed process the threat of domino effect is mitigated and in the process offers proper resource utilisation and job scheduling.

The whole process does have human involvement where the users is designated with submitting of tasks into the task queue that are waiting for dispatch. Furthermore, as discussed in the above paragraph that failed tasks are isolated from impacting other tasks however, they are not left unattended (Dorigo, & Stützle, 2019). The failed task is then re-entered into the task queue and the discussed process is repeated over and over until the operation is successfully delivered.

Particle swarm optimisation has been detailed in the above section and it had been identified that the initialisation of particles is the first step. In case of the resource scheduling the process is initiated with the initialisation of particles as well. The particles are initialised randomly, while taking in account the local optimisation (best position experienced by the particle) and the global optimisation (best position experienced by the particle) all the particles that had been determined so far in context with the trajectory with an aim to regulate the position and following direction of the particles (Pongchairerks & Kachitvichyanukul, 2016). The particles are assigned with some initial velocity so that they can pursue the direction. The decision making process of the discussed method is considered to be one of the best and it have been detailed below.

The resource to a particular task is decided by different factors. The first and foremost aspect that is considered is to check the balance of the task. If the task is in need of both the resources than the one of the two is considered and other is delayed. The next step involves inclusion of the data about the resource into the system and it is updated. Now the last updated data is the new checkpoint and based on that data next steps will be taken. The velocity of the particle also varies accordingly (Abedi et al., 2017). When a new data is added to the log based on the log the velocity of the particle is altered and it impacts their local and global optimal. The difference is then measured to determine the best location and the resource associated with that point is then allocated to the task. The same measures will be adopted for the following tasks or activities until all the tasks that are part of a broader operation is completed.

From the discussion over the discussed method it is evident that the discussed method is dedicated towards the quality, distance and direction of the activity before allocating the resource (Yuce et al., 2015). As part of the process the artificial bees are initials and spread throughout the graph that is depicting the activities and its need along with the factors that are associated with it. In the next step, the different bees who act as a scout bee evaluates the scenario and the needs of the task and the capability of the resources. The next step involves summarising of all the information regarding the task and the available or needed resources. Based on the summarised findings the crucial importance of different task is determined along with the order that should be pursued to deliver the whole operation successfully (Meng, Pan & Sang, 2018). The next step involved allocation of the resources based on the findings where the fitness of the resource decides where to which task should it be allotted and to what amount.

The above discussed measures is then repeated to determine the chain of order and the need for resource allocation. The process is continued until a satisfaction level is attained. Post attaining the satisfaction level the resources are continuously allocated to the tasks according to the finding until the process is reinitiated through some external input by the user or new data is collected from the process (Yurtkuran, Yagmahan & Emel, 2018). If new data is identified in the data log that is maintained by the algorithm to allocate the resources adequately. A number of scout bees will be deployed to determine the need for improvement and accordingly the change is made.

Conclusion

The report in discussion could be summarised to emphasise that resource scheduling is one of the most complex task that is undertaken as part of a process. However, the complexity is further increased when the process is a repetitive process as is the case in manufacturing industry where the resource allocation differs according to the need and availability of the resources. The need and availability of the resources can be extensive and to ensure the success of the process high investment of time, effort and critical thinking is expected from the managing authority. It can prove to be exhaustive and can even lead to error in the allocation. However, one of the approaches through which it can be mitigated is to have an automated learning solution.

Introduction of Machine learning is one of the solution to mitigate the discussed challenge and as part of the solution the paper have discussed swarm intelligence and how it can assist in mitigating the challenge. As part of the discussion three method of the discussed intelligence have been discussed. The discussed methods are ant colony optimisation, swarm particle optimisation and bee’s colony optimisation. All of the discussed method are crucial and have they own method s of working. The paper has detailed the role of all three of the methods and accordingly presented the way they play the role in resource scheduling. Hence, in conclusion, it would be justified to state that swarm intelligence is capable of offering its services in resource allocation and that too with a great efficiency which is capable of changing the allocation ratio with the change in need.

References

Abedi, M., Chiong, R., Noman, N., & Zhang, R. (2017, November). A hybrid particle swarm optimisation approach for energy-efficient single machine scheduling with cumulative deterioration and multiple maintenances. In Computational Intelligence (SSCI), 2017 IEEE Symposium Series on (pp. 1-8). IEEE.

Campos Ciro, G., Dugardin, F., Yalaoui, F., & Kelly, R. (2016). Open shop scheduling problem with a multi-skills resource constraint: a genetic algorithm and an ant colony optimisation approach. International Journal of Production Research, 54(16), 4854-4881.

Chen, S. M., Chen, P. H., & Chang, L. M. (2013). A framework for an automated and integrated project scheduling and management system. Automation in Construction, 35, 89-110.

Dorigo, M., & Stützle, T. (2019). Ant colony optimization: overview and recent advances. In Handbook of metaheuristics (pp. 311-351). Springer, Cham.

Du, W., Tang, Y., Leung, S. Y. S., Tong, L., Vasilakos, A. V., & Qian, F. (2018). Robust order scheduling in the discrete manufacturing industry: A multiobjective optimization approach. IEEE Transactions on Industrial Informatics, 14(1), 253-264.

Faria, P., Soares, J., Vale, Z., Morais, H., & Sousa, T. (2013). Modified particle swarm optimization applied to integrated demand response and DG resources scheduling. IEEE Transactions on smart grid, 4(1), 606-616.

Farzi, S. (2009). Efficient job scheduling in grid computing with modified artificial fish swarm algorithm. International Journal of computer theory and engineering, 1(1), 13.

Gahm, C., Denz, F., Dirr, M., & Tuma, A. (2016). Energy-efficient scheduling in manufacturing companies: a review and research framework. European Journal of Operational Research, 248(3), 744-757.

Golding, P., Kapadia, S., Naylor, S., Schulz, J., Maier, H. R., Lall, U., & van der Velde, M. (2017). Framework for minimising the impact of regional shocks on global food security using multi-objective ant colony optimisation. Environmental Modelling & Software, 95, 303-319.

Gorbenko, A., & Popov, V. (2012). Task-resource scheduling problem. International Journal of Automation and Computing, 9(4), 429-441.

Jordehi, A. R., & Jasni, J. (2015). Particle swarm optimisation for discrete optimisation problems: a review. Artificial Intelligence Review, 43(2), 243-258.

Kalinowski, K., Krenczyk, D., Paprocka, I., Kempa, W. M., & Grabowik, C. (2017). Ant colony optimisation for scheduling of flexible job shop with multi-resources requirements. In MATEC Web of Conferences (Vol. 112, p. 06018). EDP Sciences.

Kumaravel, K., & Sengaliappan, M. (2016). Performance Study of Adaptive Routing Algorithm using Swarm Intelligence. Research Journal of Advanced Engineering and Science, 1(4), 176-179.

Lama, P., & Zhou, X. (2012, September). Aroma: Automated resource allocation and configuration of mapreduce environment in the cloud. In Proceedings of the 9th international conference on Autonomic computing (pp. 63-72). ACM.

Meng, T., Pan, Q. K., & Sang, H. Y. (2018). A hybrid artificial bee colony algorithm for a flexible job shop scheduling problem with overlapping in operations. International Journal of Production Research, 1-15.

Mirjalili, S. (2019). Ant Colony Optimisation. In Evolutionary Algorithms and Neural Networks (pp. 33-42). Springer, Cham.

Ning, J., Zhang, C., & Zhang, B. (2016). A novel artificial bee colony algorithm for the QoS based multicast route optimization problem. Optik-International Journal for Light and Electron Optics, 127(5), 2771-2779.

Ozturk, C., Hancer, E., & Karaboga, D. (2015). A novel binary artificial bee colony algorithm based on genetic operators. Information Sciences, 297, 154-170.

Paikrao, P. S., & Bose, R. (2018, October). Anomaly Detection Algorithms for Smart Metering using Swarm Intelligence. In Proceedings of the 1st International Workshop on Future Industrial Communication Networks (pp. 3-8). ACM.

Pongchairerks, P., & Kachitvichyanukul, V. (2016). A two-level particle swarm optimisation algorithm for open-shop scheduling problem. International Journal of Computing Science and Mathematics, 7(6), 575-585.

Rohini, V., & Natarajan, A. M. (2016). Comparison of genetic algorithm with particle swarm optimisation, ant colony optimisation and tabu search based on university course scheduling system. Indian Journal of Science and Technology, 9(21).

Russell, S. J., & Norvig, P. (2016). Artificial intelligence: a modern approach. Malaysia; Pearson Education Limited,.

soufyane Benyoucef, A., Chouder, A., Kara, K., & Silvestre, S. (2015). Artificial bee colony based algorithm for maximum power point tracking (MPPT) for PV systems operating under partial shaded conditions. Applied Soft Computing, 32, 38-48.

Sun, J., Lai, C. H., & Wu, X. J. (2016). Particle swarm optimisation: classical and quantum perspectives. Crc Press.

Sung, T. W., Tu, C. L., Tsai, P. W., & Chang, J. F. (2017, August). Short-Term Forecasting on Technology Industry Stocks Return Indices by Swarm Intelligence and Time-Series Models. In International Conference on Intelligent Information Hiding and Multimedia Signal Processing (pp. 272-279). Springer, Cham.

Yuce, B., Pham, D. T., Packianather, M. S., & Mastrocinque, E. (2015). An enhancement to the Bees Algorithm with slope angle computation and Hill Climbing Algorithm and its applications on scheduling and continuous-type optimisation problem. Production & Manufacturing Research, 3(1), 3-19.

Yurtkuran, A., Yagmahan, B., & Emel, E. (2018). A novel artificial bee colony algorithm for the workforce scheduling and balancing problem in sub-assembly lines with limited buffers. Applied Soft Computing, 73, 767-782.

Zhan, Z. H., Liu, X. F., Gong, Y. J., Zhang, J., Chung, H. S. H., & Li, Y. (2015). Cloud computing resource scheduling and a survey of its evolutionary approaches. ACM Computing Surveys (CSUR), 47(4), 63.

Zhang, Z., Long, K., Wang, J., & Dressler, F. (2014). On swarm intelligence inspired self-organized networking: its bionic mechanisms, designing principles and optimization approaches. IEEE Communications Surveys & Tutorials, 16(1), 513-537.