Tecnomatix Plant Simulation is an industrial solution provided by Siemens Product Lifecycle Management Software Inc. which can be used to model the sophisticated manufacturing processes, material flows, shop-floor layouts, etc. Its features of object-oriented architecture, discrete event simulation, and three-dimensional display of models allow users to build realistic manufacturing models precisely and based on real-life data [1].

Their main business is to produce different types of speed transmission. The major components of one particular product or its bill of materials (BOM) is shown in Fig. 9.1.

For further study, only part of their manufacturing resource model was established in Fig. 9.2 by using the previously mentioned software. The actual manufacturing route is more complicated because it involves multiple products. To increase the usage of machines, some stations were designed flexible and were occupied by several production lines. We used this software to identify the critical production routes and to simplify this particular manufacturing system, so as to demonstrate the prototype system more clearly. To highlight the operating processes of our prototype system, only key processes will be involved in the following discussion.

The detailed information of the task is described as follows. The structure of the task can be seen in Fig. 9.3. Four separate products are included in this particular order. Each product needs to be assembled from different parts. To specify, “Product 1” requires operation “Assembly 1,” which is to assemble “Part 11” and “Part 12” together. Each part needs to be manufactured through multiple processes, which are represented by gray blocks in Fig. 9.3. The scheduled processing time of each process is listed in Table 9.1.

According to the theories discussed in Section 5.7.1, qualified manufacturing cell (MC) can undertake the decomposed subtle tasks which are above the dotted line. For example, the operation of “Assembly 2,” “Part 31,” etc. can be done by a combination of machines or at a combination of several stations. The decomposed subtle tasks, which are below the dotted line, can be assigned to certain machines in MCs. These decomposed subtle tasks are usually manufacturing processes.

In this typical manufacturing system, the fundamental manufacturing resources are designed as follows.

There are 12 manufacturing machines and 2 assembly stations, hereinafter referred to as manufacturing stations. These manufacturing machines include five lathes, four milling machines, two gear-hobbing machines, and one numerical controlled processing center. For each manufacturing station, there are two buffers, acting as a material entrance and an exit, respectively. Materials will be sent to the entrance, waiting to be processed when the machine is available; and will be put to the exit, waiting to be carried by some vehicles. Manufacturing tools, fixtures, measuring tools, and backup parts are stored in the warehouses. Some raw materials, work-in-progress parts, and finished products are also temporarily stored there. Multiple forklifts and handcarts move along the paths to fulfill the material handling tasks. The detailed information of each objects in the shop floor are listed in Table 9.2.

Similarly, RFID readers and tags are also used in the material handling system. Multiple RFID readers are installed on forklifts or handcarts. Some readers are installed at the bottom of the vehicle, as shown in Fig. 9.6A–B. These readers (or antennas) can identify the tags that are located in some key locations of the shop floor as shown in Fig. 9.6C. Thus, the locations of forklifts or handcarts are available to the material handling system. For such purposes, the narrow-band antennas are applied to limit the sensing ranges and to achieve relatively precise results. Additional RFID readers can be installed on the vehicles to achieve different goals. For example, some readers are installed where the products will be placed. These readers can capture the data of the carried products and locate the exact position of a bay inside a shelf as shown in Fig. 9.6A–B. Wide-band antennas can be applied here because they can sense for wider ranges. Products then do not necessarily need to be placed very close to the antennas in order to be perceived.

The “multitype data capturing device” in Fig. 9.6 refers to the real-time and multiple-source manufacturing information sensing system (MISS), which has been discussed in detail in Chapter 3. As shown in Fig. 9.7, by applying the technologies described in former chapters, the traditional manufacturing machines are now equipped with a processing unit, a communication module, some embedded sensors, etc.; all provided by MISS.

These candidate services are assessed by adopting the evaluation method based on GRA illustrated in Section 5.7.2. The detailed calculations are included as follows.

The weights for each evaluation indicator, which are determined by the analytic hierarchy process (AHP) in this case, are denoted as w = (0.186, 0.195, 0.205, 0.147, 0.138, 0.129)T. Then, the comprehensive evaluation matrix is obtained as:

Similarly, by evaluating other candidate services based on the previously mentioned method, their corresponding comprehensive evaluation matrices can be achieved. Here, the top three services are selected in each descending queue in terms of respective relational degrees. The evaluation indicators for all candidate services are shown in Table 9.4.

As shown in Table 9.4, a total of 3⁸ service compositions can be generated. By calculating the relational degrees of all possible compositions, the highest result is 0.7959, and the corresponding optimal composition for T is {$M{C}_1^3$, $M{C}_2^4$, $M{C}_3^5$, $M{C}_4^5$, $M{C}_5^6$, $M{C}_6^4$, $M{C}_7^7$, $M{C}_8^9$}.

By calculating the relational degrees of all service compositions, the highest one is achieved as 0.7046. Accordingly, the optimal service composition for Part 41 is {$M{M}_1^6$, $M{M}_2^5$, $M{M}_3^5$, $M{M}_4^7$, $M{M}_5^5$, $M{M}_6^2$}.

At the first stage, all the information on the real-time status of machines and stations (or the equipment agent mentioned in Chapter 8) will be collected by the capability evaluation agent (CEA), as described in Section 8.3 and Section 8.5. For each process, the corresponding machines will bid it according to their manufacturing capabilities. Then, the CEA will calculate the usage with respect to the objective function (8.1). Thus, machines with the minimum usage in the machine cell will obtain this task. These steps are repeated until all the processes are assigned to certain machines. Table 9.8 shows the results of the assignments between tasks and machines. In Table 9.8, the pair of number (x, y) of the jth row and ith column means that the jth process of Task i is assigned to Machine x and the processing time is y.

At the second stage, the real-time scheduling agent (RSA) is ready for scheduling the tasks after all the processes of the tasks have been assigned to the machines and stations. The genetic algorithm (GA) designed in Section 8.6 is applied, and the result of scheduling is shown in Fig. 9.12. Fig. 9.12A is the Gantt Chart of the task assignment. Fig. 9.12B is the generations and the fitness curve of the designed GA.

In Stage 3, which is the time for manufacturing execution, the real-time manufacturing information is constantly sensed by the production execution monitor agent (PEMA) and as inputs to RSA. In case of manufacturing exceptions, the rescheduling module will generate a new production plan according to the latest information from manufacturing environment.

To demonstrate the rescheduling process, two kinds of random exceptions are tested. The major difference between the two kinds of exceptions lies in the recovery time. Recovering time of the first kind is unknown, while that of the second type is an exact time window. As seen in Fig. 9.13A, the first exception occurred at Machine 1 and Machine 6 at time t (t = 40). In Fig. 9.13B, two kinds of exceptions from Machine 1, Machine 6, and Machine 7 also occur at time t (t = 40). All these exceptions are captured by the PEMA and reported to RSA. For the first kind of exceptions, Machine 1 and Machine 6 in Fig. 9.13A will lose the capability to take any tasks. For the second kind of exceptions, the recovery time window is considered as a load to Machine 6 and Machine 7 in Fig. 9.13B. By running through the first and the second stages, the new scheduling plan is established.

We can conclude from Table 9.13 that the combination of moving task TID₁₁, TID₁₄, and TID₁₅ has the biggest f(P, L, U) value. Consequently, these three moving tasks will be handled together by Vehicle 3. Fig. 9.14 shows the routing of TID₁₁, TID₁₄, and TID₁₅ for Vehicle 3.

Based on the previously mentioned discussions, the server will assign the optimal task to this vehicle. In this case, the assigned task is to pick up some parts at site M4, and have it transported to site M14. The current task list and the real-time information are shown on the left side of the screen as in Fig. 9.16. The planned route is also shown on the screen in green arrows. Operators should follow the instructions to reach the site for picking up products.

When the vehicle arrives at the site for picking up, materials that are to be loaded will be listed in the lower left corner of the screen. Operators may click on each items of the list to see a picture of the targeted part, and a diagram of the shelf, clearly indicating where exactly is the part within the racks on this shelf. All parts that have not been loaded will be identified as shown in Fig. 9.17. During the loading of materials, the movement of the materials can be captured by the RFID readers on the vehicle. Thus, the row of the loaded parts will turn green, so that the manager and the operators are aware of the loading progress in real time.

After loading all the required materials, the system will provide instructions to the operator to reach the next location for picking up as shown in Fig. 9.18.

When the vehicle reaches the unloading location after picking up all the required materials, the system, similarly, will tell the operator which rack should the materials be put by showing another diagram of the storage rack. Once the item leaves the trolley, the leaving event will be sensed, and this item will resume red, indicating it is no longer on the vehicle. After finishing this task (Task TID 10001), the statuses of the vehicle becomes idle again, and can obtain another task from the system.

Fig. 9.20 shows the results of comparison for the traditional material handling method and the method proposed in Chapter 6. Fig. 9.20A shows the comparison of load/no-load ratio within the four vehicles from run distance. Fig. 9.20B shows the load/no-load comparison of each vehicle. From the comparison, we can see that the no-load ratio of all the trolleys is 19.6% lower as seen on the left of Fig. 9.20A. The total run distance of all trolleys for finishing all the tasks is 668 units shorter as seen on the right of Fig. 9.20A. The run distance of each trolley without loading is 195 (Vehicle 1), 56.5 (Vehicle 2), 175 (Vehicle 3), 183.5 (Vehicle 4) units shorter as seen in Fig. 9.20B.

For example, at time t, the current information of the jobs at station B and the current information of the relevant jobs of station A, C, and D are shown in Tables 9.14 and 9.15, respectively. The notations can be found in Section 4.6.

At the station B, those operations $J_x^i(x=1\sim 10)$ will be processed in the order of $\{J_1^i,J_2^i,J_3^i,J_4^i,J_5^i,J_6^i,J_7^i,J_8^i,J_9^i,J_{10}^i\}$. The tardiness penalty per unit time of each job is given as w_1∼10 = (2,1,3,2,3,4,3,1,2,2).

In this case, three exceptions listed in Table 9.16 are designed to occur among the related upstream and downstream stations of station B.

When the exceptions occur, the due times of the relevant unprocessed jobs at station B are changed, and the queuing service will update the new start time of the affected jobs as shown in Table 9.17.

Then, the queuing service will start to requeue the order of the unprocessed jobs of the job list at station B. The initial solution and the end condition are two important decisions in the application of Tabu Search (TS) algorithm since they directly affect the utilization of the algorithm. The initial solution is generated according to the size of the due date of each job. According to the rules of the earliest due date (EDD), the initial solution proposed is suitable for the algorithm. The maximum number of iteration is set as the end condition of this proposed algorithm for the sake of simplicity. According to (4.2), the initial solution is $q_k=\{J_1^i,J_2^i,J_3^i,J_4^i,J_5^i,J_6^i,J_7^i,J_9^i,J_8^i,J_{10}^i\}$. Through the proposed algorithm, the queue is optimized. The globally optimal solution is $q_o=\{J_2^i,J_1^i,J_3^i,J_5^i,J_6^i,J_4^i,J_7^i,J_9^i,J_8^i,J_{10}^i\}$. Fig. 9.23 shows the iteration process of the TS algorithm. The algorithm finds the optimal queue whose target value is “450” at the fourth iteration. But it still searches for other solutions until it reaches the end condition. It is obvious that the new queue is better, which is “295” shorter than the target value of the old queue $(\{J_1^i,J_2^i,J_3^i,J_4^i,J_5^i,J_6^i,J_7^i,J_8^i,J_9^i,J_{10}^i\})$.

From the overview of the HTCPN model in Fig. 9.27A, there is a piece of material in in-buffer Part 111 (Place MPart111); value 0 after symbol “@” indicates its time stamp. The token in this subnet is of DATAt color type, which indicates that the job ID is a string type. When the model starts running, tokens move from one place to another, and in each place one gets a certain time stamp for each token. Tokens enter the first subnet Part 111 when the line uses the material.

Details of the six macrotransitions for each processing line are presented. For example, the macrotransition for Assembly 1 is given in Fig. 9.27B. Similarly, the detail events of the three macrotransitions for Line Assembly 1 are given as assembly process event (Transition Assembly), quality detection event (Transition QDE), and storage event (Transition SE).

As seen in Fig. 9.27C, the assembly process event is used to exemplify the model. When the assembly process should be executed according to the planned time, the PN model starts running simultaneously. At the beginning of each assembly process, the model checks the status of operators and the materials in advance. It is stated that the real-time manufacturing data is captured by RFID antennas installed at different areas. The status of tokens is updated accordingly. The tokens for an operator (Place Person) and WIPs (Place Assembly1a and Assembly1b) are also DATAt type. The token for material C is represented as (DATA, INT), where DATA is material C’s ID and INT is its quantity. After checking the status, the token for person goes into Place Attendance or Place Absent, which indicates that the person is ready or unready. If the WIPs are unready, the process needs to wait. Besides, the material C’s condition is always checked. If the number is under the threshold k, the material scheduler (Place Under Threshold) receives a token with attribute (DATA, INT) that presents the material ID and the shortage status. Then, the material is replenished accordingly. In this case, once all the operators, WIPs, and materials are ready, the process starts to operate according to the plan for assembly. Moreover, the delay time (Exception) between the trigger time and planned time can be easily calculated, and this exception event can be sent to the upper-level management system in time.

Once the HTCPN model is constructed, the production performance acquirement measures are performed to analyze the real-time production performance. Fig. 9.28 shows the results of performance analysis for several factors for the case, including quality distribution, cycle time, real-time progress, and the cost of the manufacturing process.