Author: darao’sullivan, ADI System Application Manager
In the application of industrial robots and machine tools, it may involve precisely coordinating the movement of multiple axes in a specific space to complete the work at hand. Robots generally have six axes, which must be coordinated and orderly. If the robot moves along the track sometimes, there will be seven axes. In CNC machining, 5-axis coordination is very common, but some applications will use up to 12 axes, in which the tool and workpiece move relatively in a specific space. Each shaft contains a servo driver, a motor, and sometimes a gearbox is added between the motor and the shaft joint or the end effector. Then, the system is interconnected through industrial Ethernet, generally using line topology, as shown in Figure 1. The motor controller converts the required space trajectory into a single position reference for each servo axis, and then circulates it on the network.
Figure 1. Network topology of multi axis machine tool.
These applications operate at a defined cycle time, which is usually equal to or several times the fundamental control / pulse width modulation (PWM) switching cycle of the underlying servo motor driver. In the environment shown in Figure 2, end-to-end network transmission delay is an important parameter. In each cycle, the motor controller must transmit the new position reference and other related information to each node in Figure 1. Then, enough time is left in the PWM cycle for each node to update the servo control algorithm calculation with the new position reference and any new sensor data. Then, each node applies the updated PWM vector to the servo driver at the same time point through the distributed clock mechanism depending on the industrial Ethernet protocol. According to the specific control architecture, part of the control loop algorithm can be realized in PLC. If any relevant sensor information is received on the network, it will take enough time to realize.
Figure 2. PWM cycle and network transmission time.
Data transmission delay
Assuming that the only traffic on the network is the periodic data flow between the machine tool controller and the servo node, the network delay (TNW) is determined by the number of network hops to the farthest node, the network data rate and the delay suffered by each node. In the use of robots and machine tools, the signal transmission delay caused by the line can be ignored, because the cable length is generally relatively short. The main delay is bandwidth delay; that is, the time required for data transmission to the line. For the smallest Ethernet frame (generally applicable to machine tool and robot control), please refer to figure 3 for bandwidth delay of 100Mbps and 1Gbps bit rate. This is equal to the packet size / data rate. For multi axis system, the typical data payload from controller to server consists of 4-byte speed / position reference update and 1-byte controller update of each server, that is, the payload of 6-axis robot is 30 bytes. Of course, some applications have more information in their updates and / or have more axes, in which case the packet size is larger than the minimum size.
Figure 3. Bandwidth delay of minimum length Ethernet frame.
In addition to bandwidth delay, other delay elements are due to Ethernet frames generated by phy and dual port switch of each servo network interface. These delays are shown in Fig. 4 and Fig. 5, in which the moving part of the display frame passes through phy and enters MAC (1-2). When analyzing the target address, only the preamble and target part of the frame need to be timing controlled. Path 2-3a represents the interception of the payload data of the current node, and path 2-3b represents the journey of the frame to the target node. FIG. 4A shows only the payloads transmitted to the applications in 2-3a, while Fig. 4B shows most of the transmitted frames; this indicates that there may be subtle differences between Ethernet protocols. Path 3b-4 represents the frame outbound transmission, through the transmission queue, through PHY, and then back to the cable. This path does not exist in the line terminal node shown in the figure. It is assumed that direct packet switching is used instead of store and forward. The delay time of the latter is longer because the whole frame is counted into the switch and then forwarded.
Figure 4. Frame delay: (a) dual port mode frame delay and (b) line terminal node.
Fig. 5 shows the delay elements of a frame by timeline, which describes the total transmission time of the frame through an axis node. TBW stands for bandwidth delay, TL_ 1node represents the delay of a frame passing through a single node. In addition to the delay associated with the physical transmission of bits over the line and the inclusion of address bits for target address analysis, phy and switchpack delay are other factors that affect the transmission delay in the system. With the increase of bit rate and the number of nodes on the line, these delays will have a greater impact on the end-to-end frame transmission delay.
Figure 5. Frame transmission timeline.
Low latency solutions
ADI has recently introduced two new industrial Ethernet phys designed for reliable operation in a wider ambient temperature range (up to 105 ° C) and under severe industrial conditions, with excellent power and delay specifications. Adin1300 and adin1200 are designed to address the challenges mentioned in this article and are ideal for industrial applications. With fido5000 real-time Ethernet and multi protocol embedded dual port switch, ADI company has developed a solution for deterministic time sensitive applications.
Table 1 lists the delay caused by phy and switch, assuming that the receive buffer analysis is based on the target address, and assuming a 100Mbps network.
For example, if these delays are included in a line network of up to seven axes and the total payload is included in the final node (3a in Figure 4), the total transmission delay becomes
Where 58 × 80ns represents the remaining 58 byte payload after the preamble and destination address bytes are read.
This calculation assumes that there is no other traffic in the network, or that the network can give priority to time sensitive traffic. It depends on the protocol to some extent. According to the specific industrial Ethernet protocol, the calculated value will be slightly different. Looking back at Fig. 2, when the cycle time of the mechanical system is reduced to 50 μ s to 100 μ s, the frame transmission to the farthest node may take up nearly 50% of the whole cycle, resulting in less time for the next cycle to update the motor control and motion control algorithm. Minimizing this transmission time is important for optimizing performance because it allows longer and more complex control calculations. Since the delay associated with line data is fixed and bit rate dependent, the use of low delay components (such as adin1200phy and fido5000 embedded switches) will be the key to optimizing performance, especially when the number of nodes increases (such as 12 axis CNC machine tools) and the cycle time decreases. Switching to Gigabit Ethernet can significantly reduce the impact of bandwidth delay, but it will increase the proportion of total delay caused by switches and PHY components. For example, the network transmission delay of 12 axis CNC machine tool with Gigabit network is about 7.5 μ s. In this case, the bandwidth element is negligible, and using the minimum or maximum Ethernet frame size does not make any difference. The network delay can be roughly divided equally by phy and switch. With the industrial system switching to Gigabit network speed, the control cycle time is shortened (the cycle time displayed by EtherCAT is 12.5 μ s), the number of nodes is increased due to the increase of Ethernet connected sensors in the control network, and the network topology is becoming flat, which highlights the value of minimizing the delay of these elements.
In high-performance multi axis synchronous mobile applications, the control timing requirements are very accurate, deterministic and time critical, and the end-to-end delay is required to be minimized, especially when the control cycle time is shortened and the complexity of the control algorithm is increased. Low latency phy and embedded through switch are important components to optimize these systems. To address the challenges described in this paper, ADI recently launched two new robust industrial Ethernet phys, adin1300 (10Gb / 100GB / 1000gb) and adin1200 (10Gb / 100GB). author
Darao ‘Sullivan is a senior systems application engineer in the motor and power control team (MPC) of ADI’s automation, energy and sensors business. Its area of expertise is power conversion and control for AC motor control applications. Dara holds bachelor’s, master’s and doctoral degrees in engineering from the University of cork, Ireland. Since 2001, Dara has been engaged in research, consulting and industrial and renewable energy applications.