Design of an Intelligent Shop Scheduling System Based on Internet of Things

Abstract

In order to optimize the functionality of automated guidance vehicles (AGVs) in logistics workshops, a wireless charging and task-based logistics intelligent dispatch system was developed based on the Internet of Things. This system aimed to improve freight efficiency in the workshop’s logistics system. The scheduling system successfully addressed the round-trip scheduling issue between AGVs and multiple tasks through two degrees of improvement: the application of AGVs and task path planning. To handle conflict coordination and AGV cluster path planning, a shortest path planning algorithm based on the A* search algorithm was proposed, and the traffic control law was enhanced. The initial population of genetic algorithms, which used greedy algorithms to solve problems, was found to be too large in terms of task distribution. To address this, the introduction of a few random individuals ensured population diversity and helped avoid local optima. Numerical experiments demonstrated a significantly accelerated convergence rate towards the optimal solution.

1. Introduction

An AGV (Automated Guided Vehicle), or commonly referred to as an Automated Guided Vehicle, is a self-operating transportation tool designed to perform material handling tasks without human intervention. AGVs are extensively utilized in industrial settings for automated logistics and material handling processes, with the aim of substituting conventional manual handling and logistics transportation equipment.

AGVs play a crucial role in modern logistics equipment and have gained increasing importance with the advancement of science and technology in automation. They enable automatic material distribution and transportation, leading to improved processing efficiency, reduced logistics costs, and enhanced production efficiency for businesses. A key focus of current research is the scientific deployment of AGVs, the primary equipment for logistics transportation, and the optimization of material distribution paths. By enhancing AGV scheduling strategies, sustainable business development can be achieved through reduced manual involvement, improved transportation efficiency, and increased factory production efficiency.

The scheduling of AGV clusters has been extensively investigated by specialists and academics. Various modern optimization methods, including genetic algorithms [1], ant colony algorithms [2], and particle swarm optimization algorithms [3], have been employed to minimize travel time and increase scheduling effectiveness. Yu Jiaqiao et al. [4] enhanced conventional genetic algorithms by improving coding rules, mutation operators, and crossover operators, resulting in an AGV scheduling model with the objective of minimizing the total distance traveled. This approach prevents local optimal solutions and expands the search space compared to conventional genetic algorithms. However, the technique is excessively complex. Zhang et al. [5] developed a genetic algorithm with the optimization objective of optimizing overall cost capable of simultaneously scheduling single-load and multi-load AGVs. Yu Xingbao et al. [6] introduced a simulated annealing genetic algorithm to improve the response time of the scheduling system. They created a new population using a genetic algorithm and applied the metropolis algorithm to track population state changes until stabilization was reached. Zou et al. [7] addressed the multi-load AGV task scheduling problem in a matrix manufacturing workshop by developing the discrete artificial bee colony method and the iterative greedy algorithm. These methods aim to optimize the total transportation cost. To enhance the local optimization capability and generate higher-quality solutions, Xu Fuqiang et al. [8] proposed a hybrid particle swarm optimization algorithm. They incorporated mutation and crossover operators from genetic algorithms to ensure population diversity and tested the efficiency of the algorithm. Li et al. [9] developed an improved harmony search algorithm to address multi-load AGV task scheduling in a productive shop. The optimization objectives were the total AGV traveling distance and the standard deviation of the differences between actual and required delivery times for the AGV load tasks. Liu et al. [10] introduced a multi-adaptive genetic algorithm and a mathematical model with the objectives of reducing Makespan, minimizing AGV quantity, and minimizing AGV power consumption. Their goal was to overcome these challenges. Dai Min et al. [11] proposed an improved distribution estimation algorithm with optimization goals for transit time and overall workshop energy usage. Overall, these studies demonstrate various approaches to optimizing AGV scheduling, considering different objectives and utilizing techniques such as genetic algorithms, ant colony algorithms, particle swarm optimization algorithms, and hybrid algorithms. The findings contribute to the advancement of AGV deployment strategies and the optimization of material distribution paths in logistics systems.

Wen Qiang Zou et al. [12] investigated an automated guided vehicle scheduling problem for extracting and delivering goods during processing. They proposed an effective multi-objective evolutionary algorithm to solve this problem. Yishuai Lin et al. [13] suggested a system task scheduling optimization method based on multi-load AGVs. Their optimization objectives include minimizing the number of occupied AGVs, reducing travel time, and minimizing conflicts. Wen Qiang Zou et al. [14] conducted a study on the multi-AGV scheduling problem (MAGVSCM) to optimize time and electricity costs. They proposed a mixed integer linear programming model and an adaptive iterative greedy algorithm as their solutions to this problem. Qihao Liu et al. [15] presented an integrated coding method to enhance the integration of manufacturing systems. They achieved this by simultaneously representing process routes, workshop scheduling plans, and transportation task allocation plans within a single individual. Additionally, they designed an improved genetic algorithm (IGA) to optimize this integrated system.

The objective of this article is to address the challenge of reducing labor costs and enhancing production efficiency in industrial manufacturing. To this end, we propose an intelligent workshop scheduling system incorporating the Internet of Things (IoT). This system employs the A* algorithm as a means of determining the shortest path for automated AGVs. Additionally, we propose the utilization of a traffic control method in combination with a time window strategy to overcome conflicts that may arise in the paths of AGV clusters. In terms of task allocation, the system leverages a greedy genetic algorithm to optimize task assignment. Comparative experiments were conducted between this approach and traditional algorithms to assess its efficacy.

2. Background and Scheme

The workshop faces difficulties as a result of the shift in the consumption paradigm, including an increase in dynamic events and a shorter delivery cycle. The Internet of Things-based workshop scheduling system that was developed to address this issue has to increase the handling efficiency of the workshop logistics system in order to save time and labor expenses, as well as increase the system’s stability in order to produce dependable scheduling [16]. AGV task assignment technology and AGV path planning technology are both parts of AGV scheduling technology. Figure 1 depicts the design scheme’s general layout.

3. Path Planning and Conflict Coordination Strategies

Initially, AGV path planning was investigated, as illustrated in Figure 2, which showcases the development of a polymorphic raster map centered around bi-dimensional coding. The X-axis of the chosen navigation coordinate system denotes the direction of AGV departure.

Subsequently, the A* algorithm is employed for path planning, as showcased in Figure 3, illustrating the A* algorithm’s path-finding technique. The A* algorithm, a self-heuristic search methodology renowned for its efficient algorithm for immediately acquiring the shortest path, is extensively utilized in the realms of warehousing and logistics. It possesses significant commercial landing value and outperforms other search path algorithms in terms of efficiency. The A* algorithm cycles until the AGV reaches the target by searching a suitable place in the model space, introducing the evaluation equation to assess the searched sub-solutions, choosing the best solution as the next search position, and so on. This search and evaluation process bears semblance to the concept of a “heuristic function”, which served as inspiration for the exploratory path.

$$f\left(n\right)=g\left(n\right)+h\left(n\right),$$

(1)

in Equation (1), $g(n)$ and $h(n)$ denotes “cost required from starting point to next node” and “estimated cost from node n to the target node”, respectively. The comprehensive heuristic function, denoted as $f(n)$, which corresponds to path node n, is the sum of the two and directs the subsequent search.

Three parameters $f$, $g$, and $h$ constitute this algorithm’s heuristic function [17]. The computation of $h$ utilizes the Euclidean distance, while the computation of $g$ employs the eight neighborhood search method. Equation (2) illustrates how to calculate $h$:

$$h\left(n\right)=\sqrt{(X_1-X_2{)}^2+(Y_1-Y_2{)}^2},$$

(2)

for instance, moving along X costs 10 units, while moving along Y costs 14 units. The process involves selecting the square in the “open list” with the lowest value of $f(n)$ from the “open list”, removing it from the table, and adding it to the “close list”. If the adjacent grid next to it is available, check to see if the “open list” can be added, the grid can be calculated, and the new “parent grid” can be set. Only the lowest value will be the shortest path after traversing the lowest $g$ value in the entire “open list” and becoming the new “parent square”. This computation technique can be used to determine the shortest path, and Figure 3 depicts the A* algorithm path-finding procedure.

3.1. Time Window Method

The time window method assumes that all AGVs are moving at the same speed and uniformity in order to abstract the AGV conflict problem into a mathematical model. This allows one to determine how long it will take each AGV to finish a particular path, and the time frame is then utilized for precise time scheduling. Equation (3) displays the mathematical representation of the temporal window method.

In Equation (3), $D_i$ is the time window of the i AGV, $d_i$ is the number of the grid through which the AGV path passes, $t_{iK}^{im}$ is the time when the AGV enters the grid, and $t_{iK}^{out}$ is the time when the AGV exits the grid.

Additionally, the concept of priority is employed to resolve path conflicts by making low-priority paths wait until the high-priority paths have finished their conflicting paths. The system’s programmability is increased by this strategy, but the system’s efficiency is decreased because it necessitates a fast AGV and results in pointless waiting.

$$D_i=\left\{d_i=\left[t_{iK}^{im},t_{iK}^{out}\right]\right\}.$$

(3)

3.2. Traffic Control Law

The enhanced traffic control law [18] is the conflict-coordinating technique used in this paper. The reasons for conflict are first examined. Grid conflict and cross conflict are the two sub-types of multi-Avg conflicts.

When solving grid conflicts, each AGV is assigned a pointer with an information number to indicate the current position of each AGV and the next position of the planned path to point to the grid map structure, indicating that the grid is occupied. When another AGV plans its next path grid, it determines whether the grid has a pointer to it. If the point exists, it indicates that a conflict has occurred. Due to the different AGV dispatch sequences, the scheduling system requires varying processing times to process the scheduling. The time window model formula is used to establish a strategy to resolve conflict based on the time sequence. That is, the AGV applying to the system first applies for the normal running of the point and then re-plans the path after adding the grid to the closed list. After each time the AGV travels one grid, it releases the previous pointer and then applies for the point, and the grid network map is updated every time. On paper, multi-threading is used for scheduling, but the information processing is aggregated to the total thread. Therefore, in extreme cases, when two AGVs apply for pointers from the system at the same time, the AGV that starts first will be given priority to provide pointers due to the different order of departure. When solving grid conflicts, the mathematical model of the time window is first established. According to the principle of preemption, the target grid has priority. The late arrival determines whether it is preempted and then determines whether to wait or change the path according to the mathematical model of the time window. The enhanced traffic control law is the conflict-coordinating technique used in this paper [14]. The reasons for conflict are first examined. Grid conflict and cross conflict are the two sub-types of multi-Avg conflicts.

The aforementioned techniques can be used to realize multi-AGV path planning. The enhanced traffic control system and the A* algorithm are simulated in this paper by creating the simulation platform depicted in Figure 4.

4. Design of the Task Assignment Mechanism

4.1. Greedy Algorithm

An algorithm to utilize the optimal solution is the greedy algorithm [19]. The fundamental concept behind it is to choose the best option at each stage before arriving at the overall best option. In greedy algorithms, each stage of the selection process only takes into account the present state without taking into account the overall context or previous situations. Although the greedy approach is straightforward and effective, it is frequently used to resolve both optimization and combined optimization issues but does not guarantee finding the globally optimal solution because the optimal solution chosen at each step may not always be the globally optimal solution.

4.2. Genetic Algorithm

The genetic algorithm, or GA, emulates the principles of Mendelian genetic law to discover the optimal solution by abstracting mathematical models into chromosomes or biological populations. With the rapid advancement of computational technology, researchers have shown interest in utilizing genetic algorithms to resolve a variety of challenging issues. Genetic algorithms are currently a popular AI technique for solving nonlinear issues. Without taking the function’s continuous decision, it can be operated straight from the structure and can adaptively optimize the solution. Figure 5 depicts the genetic algorithm’s flowchart.

4.3. An Improved Genetic Algorithm Based on Greedy Strategy

The greedy algorithm is employed in the genetic algorithm-based approach to expedite computation. The genetic algorithm is the fundamental factor in the process of finding the optimal solution. The traditional genetic algorithm’s population initialization typically involves computer-generated randomness, making it easy for the initial population’s individual fitness to differ significantly from the optimal fitness we seek. As a result, the algorithm’s convergence speed is slow and its running time is long, which lowers the quality of the solution. In this paper, the greedy algorithm is used to generate generation and randomly generated generation, and part of the generation is generated by the greedy algorithm and the other part is generated randomly by the computer, which improves the quality of the initial population, ensures the diversity of the initial population, improves the initial convergence speed of the generation, but also prevents the local optimal.

4.4. Algorithm Implementation

4.4.1. Improved Genetic Algorithm Coding Design

Binary code, gray code, and floating-point code are frequently utilized in the coding design of the conventional genetic algorithm. However, in this study, breakpoint-based natural number coding is utilized. Points uniquely establish where each route begins and finishes by providing randomly generated points to show where one route ends and the next begins. Additionally, the AGV uses the chromosome sequence to determine the city number of the gene. Assuming there are three sets of AGV, the breakpoints for 1 and 10 are [2, 5, 7, 1, 12, 9, 6, 8, 10, 11], respectively. Routes for AGVs 1 and 2 are [0-2-5-7-1], [0-12-9-6-8-10], and [0-11-3-4-0].

4.4.2. The Formation of Initial Populations

The initial generation has a profound impact on how well a genetic algorithm performs when it is used to resolve the MTSP (multiple traveling salesman problem). In order for the algorithm to attain the overall best solution, we generally need to produce a suitable set of beginning populations. The starting population, which was chosen randomly, might not be able to satisfy this criterion. Therefore, we may construct roughly 20 ideal people using the greedy technique to avoid local optimal solutions being formed because of too rapid convergence. To add more randomization, the initial generation was improved with forty artificially generated individuals.

4.4.3. Fitness Functions and Selection Operations

Path optimization seeks to evaluate the fitness of the coding chromosome in the population by building the fitness function of the genetic algorithm using the path length mapped by the coding chromosome. Assume that P, which has m pathways, is the multi-travel salesman problem solution. For each path, we assume it contains cities, so its length is. The fitness function of the whole solution can be expressed as Equation (4), shown below:

$$fitness\left(p\right)={\sum }_{i=1}^{m}dis{t}_{Pi},$$

(4)

$$dis{t}_{Pi}={\sum }_{j=1}^{n_i-1}dist(P_{i,j},P_{i,j+1})+dist(P_{i,n_i},P_{i,1}),$$

(5)

in Equation (5), $dist(p_{i,j},p_{i,j+1})$ represents the distance from city $p_{i,j}$ to $p_{i,j+1}$, and table $dist(p_{i,n_i},p_{i,1})$ is the distance from the start to the end of path $p_i$.

The tournament selection approach was used in the selection process, where the person with the highest fitness evaluation was selected and added to the progeny population based on the fitness value of each individual. In this method, the population is randomly divided into various groups at random using the concept of a tournament, and each group is then evaluated by the adaptation function, which creates a new population with the specified proportion.

4.4.4. Crossover and Mutation Operators

The population solutions presented in this paper, generated through a genetic algorithm and stored in arrays, contained the job assignment solutions for different AGVs arranged in a specific order. Three continuous solutions and two continuous solutions from the parent algebraic group are randomly exchanged in the crossover process. The self-exchange of the sub-generation array produced by extracting the variation ratio with its random three continuous solutions and two continuous solutions is known as variation.

4.4.5. Greedy Operator

The given city is encoded to obtain a set of ordinal numbers. Then add a randomly generated breakpoint. Where the evaluation function is represented as:

$$f(i,j)=\sqrt{(x_i-x_j{)}^2+(y_i-y_j{)}^2}$$

(6)

by employing a randomization process, an initial city i is found, followed by a city j that is closest to city i in the remaining cities to form a path, determining whether it is the shortest distance by Equation (6), and selecting the city that is the shortest distance from the last selected city in the remaining cities in turn until a complete path is formed. In this way, the initial population formed by the city path has good fitness at the beginning, which is helpful to the convergence of the algorithm and the improvement of the solution accuracy.

5. Experiment and Result Analysis

Based on the genetic algorithm of the greedy strategy established in the above section, a program simulation was written. The hardware processor simulated in this paper is the AMD Ryzen 7 4800H with Radeon Graphics at 2.90 GHz, which runs the 64 bit operating system Windows 10. The simulation software is MATLAB. Before the simulation, the mathematical model established in this paper is instantiated; it is assumed that the task cycle of 28 stations is applied to the scheduling system. The 27 stations are distributed in the two-dimensional coordinate plane of (−100, 100), the regional logistics center is located at the global map coordinate origin of the park (0, 0), the starting point of AGV is set at point 1, and the traffic network between each point is fully connected, as shown in

Table 1. Location information for each station.

Number	Coordinate	Number	Coordinate
1	(0, 0)	15	(−48.98, 81.84)
2	(−98.23, 97.87)	16	(−63.21, 34.83)
3	(48.74, 26.84)	17	(−87.72, 25.83)
4	(15.56, 56.38)	18	(−63.23, 34.16)
5	(−08.78, 96.84)	19	(−62.12, 7.18)
6	(−63.84, −21.82)	20	(98.67, 97.45)
7	(94.34, 50.65)	21	(87.31, 13.37)
8	(61.63, 87.28)	22	(78.93, −53.23)
9	(33.79, −8.84)	23	(86.45, 80.21)
10	(−83.21, −10.34)	24	(70.24, 46.23)
11	(−93.23, −71.26)	25	(−37.45, −39.21)
12	(54.23, 72.47)	26	(−30.21, −68.21)
13	(11.25, 10.85)	27	(14.24, −48.87)
14	(31.54, 21.46)	28	(−9.51, −91.85)

. In this paper, the population number N is 60, the mutation rate is 0.6, the number of AGVs is 4, and the number of iterations is 1000 generations.

Figure 6 shows the iterative diagram of the traditional genetic algorithm and the iterative diagram of the improved genetic algorithm. Figure 7 shows the traditional and improved genetic algorithm task assignment diagrams. Table 1 and

Table 2. Optimal AGV scheme distribution route.

AGV Number	Handing Shelf Order
1	1−19−17−18−3−16−6−5−1
2	1−20−11−7−12−26−27−28−1
3	1−13−9−21−24−14−1
4	1−10−23−22−8−25−4−15−1

show the location information of all stations and the optimal AGV distribution route.

In terms of the number of iterations, it can be seen in Figure 6 that when four AGVs are scheduled to complete 27 tasks, the improved genetic algorithm based on the greedy strategy in this paper has significantly fewer iterations than the traditional genetic algorithm. The traditional genetic algorithm has not calculated the optimal solution when iterated over 1000 generations, but the improved genetic algorithm has obtained the optimal solution after 260 iterations. Regarding distance optimization, the analysis of Figure 8 and Figure 9 demonstrates that with an equal number of iterations, the total driving distance of AGV in the improved genetic algorithm is 1331.5453, while that in the traditional genetic algorithm is 1397.8703. The results of the improved genetic algorithm are closer to the optimal solution. In terms of calculation time, the average calculation time of the traditional genetic algorithm measured by MATLAB is 1.95 s, and the average calculation time of the improved genetic algorithm in this paper is 0.97 s, which is significantly optimized compared with the traditional algorithm.

6. Scheduling Platform Construction

To make the dispatching terminal and multiple AGVs have reliable wireless long-distance communication, this design uses WIFI for communication, and its transport layer protocol uses the TCP protocol, which is reliable communication based on a byte stream. The host computer uses socket technology to receive data and multi-threaded technology to receive information from the AGV cluster. After each schedule, the information is inserted into the database. The database used in the design is SQLite, and the host computer for scheduling is shown in Figure 8.

The dispatch system is divided into several key components, namely the network configuration area, map configuration area, interactive display area, vehicle configuration area, and database query area.

The map configuration area serves as the initial setup for the grid map. Users can configure the number of grids to be displayed in the table box below. Additionally, the scheduling results are visualized in this area. In the network configuration area, users have the ability to configure the IP address and port number of the server. It also offers the option to either listen or stop listening for incoming connections. The interactive display area provides a visual representation of the AGV connections, the data being sent, and the respective ID numbers of the AGVs. The vehicle configuration area is used to initialize the A* algorithm and scheduling settings. After receiving the positioning data from the AGV, users can set the starting point. Once the task points are set, clicking the departure button will initiate the scheduling process.

Lastly, the database query area enables users to search and visualize the database. Clicking the “Schedule Data View” button opens a separate window to display the queried database, as shown in Figure 9.

To verify the feasibility of the scheduling algorithm, the hardware structure was set up in this paper, as shown in Figure 10. The motor speed measured by the encoder depends on the motor change gear’s feedback to the encoder speed measuring gear to measure the speed, and the gear transmission ratio is 3:10.

7. Conclusions

In this paper, an intelligent scheduling system for workshop logistics based on the Internet of Things (iota) integrated wireless charging and tasks is implemented. By improving or applying AGV’s path planning and task assignment, the round-trip scheduling problem between multiple tasks is effectively resolved. At the level of path planning, this study proposes shortest path planning based on the A* pathfinding algorithm and realizes multi-AGV conflict coordination and path planning by improving traffic control law. At the task allocation level, this study solves the problem of a too long iteration time for the genetic algorithm by using a greedy algorithm to generate part of the initial individuals of the genetic algorithm and add part of random individuals to ensure population diversity while preventing local optimality. Numerical experiments have demonstrated that the traditional genetic algorithm yields the shortest path of 1397.8703, whereas the greedy genetic algorithm produces a shortest path of 1331.5453. The average execution time for the traditional genetic algorithm is 1.95 s, while the greedy genetic algorithm takes an average of 0.97 s. The algorithm proposed in this article ensures an optimal solution, significantly enhances the convergence speed, and exhibits high practicality and feasibility.

Considering the rapid advancement of industrial automation and the increasing complexity of scheduling requirements in the future, there will be a greater demand for improved scheduling efficiency. Future research can focus on tailoring solutions to specific transportation scenarios and dynamic AGV task requirements. In terms of optimization methods, the heuristic evaluation function can be enhanced to filter solutions more effectively, or a hybrid algorithm that combines multiple heuristic algorithms can be devised. This article solely presents a theoretical solution for scheduling technology applications. To meet production needs in practical workshop environments, targeted design and secondary development are necessary, taking into account the specific characteristics of different workshop settings.