以硬體實現主動可調式CAN網路排程系統

Abstract

在本論文中分別使用Borland C++ Builder建立CAN（Controller Area Network）網路系統的模擬軟體，並建立8051與SJA1000 Standard CAN Controller組成多組CAN節點的硬體實驗系統，實驗依據三項傳輸指標：平均傳輸時間、未即時傳遞訊息比例以及訊息遺失比例，在不同的負載量下，分析不同排程理論的優劣。 因為Rate Monotonic scheme與Deadline Monotonic scheme無法在訊息傳輸的過程中更改訊息的優先權，使得優先權低的訊息在高負載量時無法即時傳送，所以本研究希望在訊息傳送的過程中，可以即時更改訊息的優先權以提升傳輸效率，文中使用動態調整訊息的理論有Earliest Deadline First scheme以及本論文研發出的Active Adjustable Priority scheme。EDF是訊息快要錯過deadline的時候，即時調整訊息的優先權，使得訊息可以快速送出。而AAP是累積訊息一段時間的傳輸狀況後逐漸調整優先權。EDF因為硬體SJA1000的設計限制，無法在硬體上實現，而AAP可以在硬體上實現。 對於在硬體實驗上顯示對於missing deadline percentage來說，DM＋AAP有很顯著的效果。而lost message percentage則以RM＋AAP的效果最好。對於網路控制系統而言，如果訊息是作為回饋控制，未能及時傳遞時，會造成系統不穩定，在這種系統架構下選用DM＋AAP最能達成系統穩定。而訊息若是作為控制命令，控制命令如果遺失，系統無法做出回應動作造成系統有錯誤發生，因此選用RM＋AAP可以使系統發生錯誤的機會最少。

By using the Borland C++ Builder software, a controller area network (CAN) bus simulation system is developed in this thesis. Furthermore, results of a hardware system with different bus loads and nine CAN nodes, which combine 8051 with SJA1000 standard-alone CAN controller, are compared with simulation results in three indices: (1) the average transmitting time, (2) the missing deadline percentage, and (3) the lost message percentage. Simulation and experimental results indicate that the rate monotonic (RM) scheme and the deadline monotonic (DM) scheme, which are formed as the fixed message priority during the period of transmission, successfully transmit the higher priority messages only in time. Therefore, to develop real-time changing priority of CAN messages in this thesis leads to two adjustable schemes: (1) earliest deadline first scheme (EDF) and (2) the proposed active adjustable priority (AAP) scheme. Theoretically, as messages are going to miss their deadlines, the EDF changes their priorities immediately while the proposed AAP gradually changes its priority according to performance of message transmission. Physically, the EDF cannot be implemented on real hardware system and the AAP is adopted in the present experiments. Experimental results also indicate that the DM+AAP presents the best performance in the missing deadline percentage and the RM+AAP presents the best performance in the lost message percentage. For network control systems, the deadline for the sensors to provide the measurement is critical for performance and stability in feedback control. Therefore, the DM+AAP scheme is more suitable as bus message load increases. On the other hand, as messages of commands for each CAN node is more important, like in the synchronization for multi-axis systems, the lost message is become more critical in contributing the errors. Thus, adopting the RM+AAP scheme is recommended.