Design of Electro-hydraulic Position Servo Control System Based on DSP NNC-PID

In the process of automobile manufacturing, a large number of electro-hydraulic position servo manipulators (welding, painting), machine tools (stamping, pressing) and other processing devices are used. The electro-hydraulic position servo system has the characteristics of high power, fast response and high precision, which requires the control system to have not only good positioning accuracy, but also good servo tracking performance, so it is an important part in the control field. The typical characteristics of the electro-hydraulic position servo control system are nonlinearity, uncertainty, time-varying, external interference and cross-coupling interference, etc., and it is not easy to establish an accurate mathematical model of the system. Therefore, the control of the electro-hydraulic system has always been a complex control system problem.
Conventional PID controllers have the characteristics of simple structure, clear parameter meaning, and excellent dynamic and static characteristics of control. Artificial neural network (NNC) has the ability of information integration, learning memory and adaptiveness, and the ability to approximate arbitrary nonlinear functions. It can handle processes that are difficult to describe with models and rules, but there are also local minimum points, and it is not easy to achieve optimal control.
Combining NNC with PID control to form an intelligent controller can achieve better control effects. Here, it is proposed to use DSP to realize NNC-PID controller to intelligently control the electro-hydraulic position system to meet the requirements of electro-hydraulic position servo for fast response and high precision of the control system.

1 Composition of electro-hydraulic position servo system
Taking the first joint of the spray painting robot as the object, a research experimental device was constructed, as shown in Figure 1. The feedback device uses a precision conductive plastic potentiometer. The entire control system is based on DSP and consists of the first joint of the spray painting robot, position sensor, 12-bit A/D converter and D/A converter, signal conditioning circuit and output amplifier drive circuit, as well as host computer PC, to achieve positioning and servo tracking control.

2 Control System Hardware Design TMS320F2812 is a 2000 series digital signal processor (DSP) launched by TI, mainly used in the control field. The frequency reaches 150 MHz, the fixed-point 32-bit CPU can run 16×16 and 32×32 operations. The on-chip program memory is up to 128 KB, ROM is 128 KB and SARAM is 18 KB, the external interface has 16-bit data lines and 19-bit address lines, and the ROM can be expanded to 1 MB. In addition, it also integrates a 16-channel 12-bit A/D converter with a minimum cycle of 80 ns, and 56 individually programmable general-purpose I/O (GPIO) pins. The high-speed digital signal processing capability and rich external expansion resources make TMS320F2812 suitable for control systems with higher requirements. 2.1 Overall structure of the control system The control system adopts the PC+DSP control solution, and the overall structure of the system is shown in Figure 2. The PC is mainly used to display the control interface, adjust various control parameters, and display relevant signals in real time. The DSP completes the low-level control function, collects various signals through the A/D converter, and outputs them through the D/A port after certain algorithm processing, and controls the switch of each solenoid valve through the I/O port and the photoelectric isolation drive amplifier circuit. At the same time, through communication, it sends the collected signals to the PC and receives the start, stop and other instructions and various control parameters from the PC.

2.2 A/D conversion circuit
The A/D converter module of TMS320F2812 can reach a clock of 25 MHz, with a conversion accuracy of 12 bits. It can collect 16 channels and 0-3 V voltage analog signals. There are multiple triggering modes: software trigger (DOC), event manager A (EVA), and event manager B (EVB). The relationship between its conversion data and input voltage is: digital quantity = 4 095x (V input-VADCLO)/3, where VADCLO is the reference voltage of each channel.
When wiring the PCB, the distance from the signal input end to the TMS320F2812 pin should be as short as possible. At the same time, each channel should be away from the digital signal and a large area should be laid. In the A/D converter circuit module, J3 is connected to the sensor, and J19 can be connected to an oscilloscope, etc., which can be used for data collection by other instruments.
2.3 I/O and drive design
The I/0 board is mainly used to drive each solenoid valve. The drive current can reach several amperes, and the electromagnetic noise is large. The switch of each relay will produce strong electromagnetic interference, and the current impact and voltage peak of the switch are large, which will affect the operation of the DSP. Therefore, the board is made separately from the DSP main board. In the I/O board design, 74LS244 is used as the driving element, and TLP521 is used as the photoelectric isolation and relay to drive the external load. When PCB wiring, the wires with large currents are appropriately thickened. The board can drive 8-way (expandable to 16-way) solenoid valves.
2.4 Communication circuit
The ISPl581 used in the USB communication circuit design is a universal serial bus interface device of Philips, which fully complies with the USB2.0 specification. It supports the self-test working mode of USB2.0 and the return working mode of USBl.1, directly connects to ATA/ATAPI peripherals, integrates 8 K bytes of multi-structure FIFO memory; high-speed DMA interface: 7 OUT endpoints and a fixed control IN/OUT endpoint. Through a high-speed universal parallel interface, ISPl581 provides high-speed USB communication capabilities for systems based on microcontrollers/microprocessors. Using existing structures and reference firmware not only shortens development time, but also reduces development risks and costs. It is a simple and economical USB peripheral solution.

Map ISPl581 to the XINTF ZoneO space of TMS320F2812 , use as the chip select signal of ISPl581, select a GPIO pin of TMS320F2812 as the signal to reset ISPl581, connect the read and write control signals directly, and connect the interrupt signal that plays an important role in the operation of ISPl581 to XINTl of DSP so that DSP can handle the communication interruption of USB in time. Since the storage space of ISPl581 is 8-bit, and the storage space of TMS320F2812 is 16-bit, its data lines DO~D15 can be directly connected, the address line A0 of ISP1581 is grounded, A1 is connected to A0 of DSP, A2 is connected to A1 of DSP, and so on until A7 is connected to A6 of DSP. The working mode of ISP1581 is selected as the general processor mode, that is, the separate address lines AO~A7, the processor and DMA share the data lines D0~D15, and the read and write mode is selected as the 8051 mode, that is, the read and write control is . Connect the MODE1 pin directly to +5 V and the ALE/AO pin to ground. 2.5 External memory circuit TMS320F2812 maps the external storage space into five 16-bit areas, XINTF Zone0~XINTF Zone2, XINTF Zone 6 and XINTF Zone7. Among them, XINTF Zone0 and XINTF Zone1 are both 8 KB and share the chip select signal ; XINTF Zone2 is 521 KB and the chip select signal ; XINTF Zone6 is 521 KB and XINTF Zone7 is 16 KB and share the chip select signal . The memory circuit uses the storage space of XINTF Zone2 and INTF Zone6, and selects IS61LV25616 as the storage device. The data lines D0~D16, address lines AO~A17, and read/write control lines of TMS320F-2812 and IS61LV25616 are directly connected. The A18 of TMS320F2812 forms a chip select signal through a decoding circuit composed of logic gate devices 74AC04 and 74LVC32 , thereby realizing the read/write control of IS61LV25616. 3 Neural Network NNC-PID Controller Neural network is a highly nonlinear ultra-large-scale continuous-time dynamic system with large-scale parallel distributed processing, high robustness, adaptability and learning association capabilities. It can adapt to environmental changes well and self-learn to modify process parameters. These characteristics provide great potential for the application of neural networks in the control of electro-hydraulic position servo systems. 3.1 Structure of Neural Network PID Control System The structure of the neural network PID control system is shown in Figure 3(a). As can be seen from the control system block diagram, the neural network PID control includes two control submodules: NNI is the controlled object model identifier, and NNC is the neural network PID controller. The working principle of the NNC-PID control system is: first obtain the input and output sample pairs of the actual controlled object, then use NNI to perform offline identification of the controlled object, and when the identification accuracy reaches the set requirements, adjust the weight coefficients of NNC in real time to make the coefficients adaptive, thereby achieving the purpose of effective control.

3.2 Neural Network Identifier (Controlled Object Model Identifier NNI)
The neural network identifier NNI is implemented using a three-layer series-parallel BP network, including an input layer, a hidden layer, and an output layer. Its structure is shown in Figure 3(b). The input of the network is the input/output sequence [u(k), y(k)] of the controlled object, and the output of the network is the teacher signal .
The input and output of the network hidden layer are:

3.3 Neural Network NNC-PID Controller (Single Neuron Adaptive NNC-PID Controller)
Since the controlled object model is uncertain and unknown, and there are external random disturbances, in order to achieve higher control accuracy, a single neuron adaptive NNC-PID controller structure is adopted based on the offline identification of the controlled object model, as shown in Figure 4.

The network weight coefficient value V=[v1, v2, v3], that is, the three coefficients KP, KI, KD of the PID controller. The network input is X=[x1, x2, x3], that is, the three input parameters e(k), △e(k), △2e(k), and the network output is △u(k).

The supervised Hebb learning rule realizes the adaptive and self-organizing functions by adjusting the weight coefficients. The control algorithm and learning algorithm are shown in equations (10) and (11).

According to the supervised Hebb learning rule, the weight coefficient is adjusted according to the rules of equations (12) to (14) as follows:

Where K is the neuron proportional coefficient, ηI, ηP, and ηD are the learning rates of integration, proportion, and differentiation respectively.

4 System software design
The system software design is mainly divided into two parts, the PC program written in Labview and the DSP program written in C language. The PC program is used to display and process the data sent by DSP, and send instructions and adjust parameters to DSP. The system software
design of DSP is designed and written in C language under the development system of CCS2000. It adopts a top-down design idea and divides the software modules according to functions. The system software is shown in Figure 5, which mainly consists of initialization module, fault diagnosis, USB communication module, manipulator NNC control learning module and manipulator NNC-PID control module.

5 Experimental results
Conventional PID control is first used for the electro-hydraulic position servo manipulator system, and the PID parameters are adjusted using the Ziegler-Nichols method, that is, the control system is under pure proportional control, and the proportional gain is adjusted to make the system reach critical stability. The gain ku and critical oscillation period Tu at this time are recorded to determine the PID parameters, namely: kp=0.6Tu, kI=0.5Tu, kD=0.25Tu, and finally the proportional, integral, and differential coefficients are determined as: kP=1.02, kI=0.024, kD=0.006, respectively. At this time, the position step tracking response of the coefficients is shown in Figure 6. Under the same conditions, the neural network NNC-PID control method is used to control the electro-hydraulic position servo manipulator system. The initial weight of NNC is taken as the setting value of PID, that is, v1(0)=1.02, V2(0)=0.024, V3(0)=0.006. In order to ensure the stability of iteration, the iteration range of weight is limited to: 0.1≤v(1)≤1.3, 0.001≤v(2)≤0.06, 0.001≤v(3)≤5. At this time, the position tracking response curve of the system is shown in Figure 6. By comparison, it can be seen that the neural network NNC-PID method has the learning ability to make the system converge to the steady-state value of the position quickly. The neural network NNC-PID control can improve the control performance of the system because it can adjust the PID parameters in real time. At the same time, it shows good robustness to the time-varying parameters, which well solves the nonlinearity and time-varying parameters of the hydraulic system.

It should be noted that the selection of neuron proportional coefficient K has the most important influence on the control performance of the system. Too large or too small will lead to poor system performance, and even fail to achieve self-optimization and self-adaptation. The influence of ηP, ηI, and ηD on the performance of the system is reflected in the speed of learning.

6 Conclusion
By analyzing the characteristics of the operation and debugging of the electro-hydraulic position servo manipulator and its requirements for the controller circuit, a control scheme of PC + DSP based on neural network NNC-PID controller is adopted to design the hardware and software of the electro-hydraulic position servo PC + DSP control system, and the functions, characteristics and plate making requirements of each hardware control subsystem are analyzed in detail, and the controller software design process based on neural network NNC-PID and the compilation and debugging of the software are explained. After laboratory comparative operation, the control effect of the electro-hydraulic position servo manipulator PC + DSP control system based on neural network NNC-PID controller is good, the controller works reliably, and the parameters are easy to adjust.