Abstract: Based on the introduction of LONWORKS fieldbus technology, a condition monitoring system for hydro-turbine units based on LONWORKS fieldbus was designed and implemented. This system monitors various parameters of the operating hydro-turbine unit online, providing a solid foundation for comprehensive assessment of the unit's status, allocation of potential faults, and implementation of condition-based maintenance. The system structure diagram, main hardware design, message transmission protocol, system communication software design, and fault diagnosis examples are presented. Keywords: Hydro-turbine generator unit, condition monitoring, LONWORKS, neural network chip. Hydro-turbine units are crucial components in power networks. Ensuring the normal operation of large hydro-turbine units, monitoring their operating status, and promptly detecting fault signs to achieve "prevention-oriented maintenance" is a long-cherished ideal in the engineering field and a development direction for the maintenance of large power plant units. Real-time condition monitoring can reduce unit downtime and improve utilization. The condition monitoring discussed here is actually the real-time online monitoring of numerous parameters of the hydro-turbine unit. Hydro-turbine units have many parameters, and for ease of analysis, some parameters require high-speed sampling. Thus, a monitoring system typically consists of multiple acquisition nodes distributed at different field locations. Each node transmits a large amount of collected data to the host computer, which then evaluates the unit's operating status from multiple perspectives. Integrating the entire monitoring system using a fully digital communication fieldbus enables completely decentralized control, resists various interference factors, simplifies the system structure, and improves data transmission efficiency. Therefore, this paper designs and implements a turbine unit status monitoring system based on the LONWORKS fieldbus. 1. Technical Characteristics of the LONWORKS Fieldbus The LONWORKS fieldbus is a fieldbus technology launched by Echelon Corporation in the United States. It features openness, high speed, and interoperability; it adopts an object-oriented design method, simplifying network communication design to parameter settings and reducing development difficulty; it supports multiple transmission media, with a network capacity of up to 32,000 nodes, a network communication rate of up to 1.25 Mbps/130 m, and a direct communication distance of up to 2700 m/78 kbps; its network employs measures suitable for industrial environments, such as configuring a 1500V DC isolation transformer for isolation, giving it strong anti-interference and anti-vibration capabilities, making it suitable for harsh industrial environments such as hydropower plants. In the turbine unit condition monitoring system, there are dozens of field data acquisition nodes. These nodes transmit sampling results to a distant host computer via fieldbus, and data communication is frequent. Considering the strong signaling capability of LONWORKS bus technology, using LONWORKS to form the system's real-time data communication network greatly simplifies the system's communication software and hardware design, making data transmission and communication very convenient. 2 System Composition The turbine unit condition monitoring system based on LONWORKS fieldbus consists of a host computer and multiple field monitoring units, as shown in Figure 1. Each field monitoring unit monitors three field data acquisition nodes. The field monitoring units of the field data acquisition nodes are mainly responsible for collecting, storing, and transmitting field data. Field monitoring includes pressure monitoring; temperature monitoring; water level and oil level monitoring; water flow monitoring; unit vibration and swing monitoring; unit electrical monitoring; unit speed and guide vane opening monitoring; efficiency monitoring; turbine cavitation monitoring; generator air gap monitoring; generator insulation monitoring; and tailrace vacuum monitoring, etc. Each field acquisition node forms a field monitoring network via the LONWORKS bus. A developed LONWORKS-ETHERNET interconnect adapter converts the data sent by the acquisition nodes on the LON bus into UDP format, then transmits it to the upper-level computer via high-speed industrial Ethernet. Commands from the upper-level computer are then converted into LonTalk protocol format and sent to each field node, thus enabling communication between the upper-level computer and the underlying field nodes. Compared to using expensive LON bus adapters in short training courses, this method improves data transmission speed and saves on costs. The upper-level computer stores the various monitoring data transmitted from the field nodes into an MS-SQLSERVER2000 database, provides a human-computer interaction interface, and performs graphical and formatted display of real-time data. It also analyzes the data using Fourier transform (FT) and wavelet transform (WT). 2.1 Field Node Design The field nodes must receive acquisition commands from the upper-level computer, instructing standard sensors to acquire field signals; they must also send the acquired field signals to the upper-level computer via the LON bus for processing. Their structure is shown in Figure 2. 2.1.1 Node Composition The node consists of a Neuron 3150 neuron chip, LONWORKS twisted-pair cable, network transceiver, programmer, data memory, and a 12-bit A/D converter chip AD1674. Specifically, the Neuron 3150 neuron chip is the TMPN3150 manufactured by TOSHIBA; the FLASH ROM is the AT29C512; the data memory (RAM) is the IS61C256 from ISSI; the network interface between the Neuron 3150 chip and the LON bus uses the Echelon FTT 10A free-topology transceiver, a transformer-coupled transceiver that provides an inaccessible interface to the twisted-pair cable and supports free network topologies; the network communication medium uses the most commonly used twisted-pair cable; the A/D converter chip is the cost-effective AD1674 chip, with a conversion accuracy of 1/2 LSB, a conversion rate of 100 kSPS, and a tri-state output buffer. 2.1.2 Memory Space Allocation The on-chip memory of the Neuron 3150 chip ranges from E800H to FFFFH, including 2KB of RAM (E800～EFFF), 0.5KB of EEPROM (F000～F1FF), 2.5KB of reserved space (F200～FBFF), and 1KB of space for memory-mapped I/O (FC00～FFFF). The addresses of external extended memory are determined by the address and control pins of the Neuron 3150: the address space allocated to the FLASH ROM is 0000～7FFF, of which 16KB of space from 0000 to 3FFF is used for system firmware (which implements the LonTalk protocol), and 16KB of space from 4000 to 7FFF is used for user program code; the address space allocated to RAM is 24KB from 8000 to E7FF; and the 2KB address space from E000 to E7FF is allocated for memory-mapped I/O of external devices. 2.1.3 A/D Conversion Interface Scheme This paper considered using the Neuron chip to provide serial and parallel I/O interfaces for the A/D conversion circuit. However, the serial I/O method is too slow; the parallel I/O method requires occupying all 11 I/O interfaces of the Neuron chip, and also requires programming the handshake/token passing algorithm of the Neuron chip, which is costly and difficult to implement. Therefore, this paper allocates the 2KB address space reserved by the node from E000 to E7FF to the A/D conversion chip as the port address of AD1674, and uses the memory mapping method to directly read the data from AD1674. For this design, the high 8 bits of the AD1674 conversion data address is E002H, and the low 4 bits of the data address is E003H. Because the software implementation does not use Neuron C's built-in functions, the execution speed is significantly improved. Experiments show that for nodes using the same AD1674 conversion chip, the node designed using this method has a higher acquisition speed than nodes designed using the other two methods, and it also saves all 11 I/O pins of the Neuron chip. 3 System Communication Program Design 3.1 Field Node Communication Program Data sent from the field node to the host computer is first sent to the LONWORKS-ETHERNET interconnect adapter. This adapter is actually a special LONWORKS node that encapsulates the received data from the LON bus using UDP and then sends it to the host computer via Ethernet. There are two main communication methods between nodes in the LONWORKS network: network variables and explicit messages. Using network variables eliminates the need to consider message packaging, sending, and receiving, which greatly simplifies programming and shortens the application development cycle. However, the data length of each cycle variable cannot be changed once determined, and it is limited to a maximum of 31 bytes. The data length of explicit messages is flexible and variable, with a maximum of 228 bytes, but the implementation method is more complex. Given the high real-time requirements of the turbine unit condition monitoring system for data transmission, and the need to improve the upload speed of the data collected from the football field, it is desirable for each transmitted message to contain as much data as possible. Therefore, the design adopts a display message method to achieve communication with the host computer. Each display message carries 134 bytes of data, of which 128 bytes are the transmitted data and the remaining 6 bytes are additional information. The frame structure of the message is shown in Figure 3. The implementation program for the initialization and sending of the display message is as follows: Initialize node address #include Required header files #include #include domain_struct mydomain; // Define the domain structure mydomain = * (access_domain(0)); // Read the node domain table mydomain.subnet = 0; // Set the node subnet number mydomain.node = 5; // Set the node number update_damain(&mydomain, 0); // Write to the node domain table Send data packets msg_tag test_out; // Declare the packet tag msg_out.tag = test_out; // Transmit the packet tag msg_out.dest_addr.snode.node = 0; // Define the destination address node number msg_out.dest_addr.snode.subnet = 0; // Define the destination address subnet number msg_out.code = 0x0c; // Define the packet code msg_out.service = ACKD; // Define the packet service type msg_out.dest_addr.snode.type = 1; // Define the destination node type memcpy(msg_out.data, a0, nLength); // Fill in the message content msg_send(); // Send the message In the initialization program, the subnet number and node number of the node are defined using the data structure domain_struct, that is, the address of the node on the LON bus is set. In the sending program, the message is constructed using the msg_out structure, where the destination address points to the adapter. The receiving program for display messages is similar to the sending part, and will not be described in detail. 3.2 Host Computer Communication Program Data communication between the host computer and the field nodes is forwarded through the adapter, and the IP of the adapter must be specified in advance. The host computer uses the msg_out variable (as described above) to create a display message, fills the target node address, the parameters to be changed or the command to be issued into this variable, and then encapsulates the variable with UDP and sends it to the adapter via Ethernet; the adapter parses the data packet sent by the host computer, obtains the display message, and sends the display message directly to the corresponding node. Similarly, the adapter encapsulates the display messages from the field nodes via UDP and sends them to the host computer. The host computer unpacks the messages and stores the data in the corresponding database based on node information, awaiting subsequent calls from the signal processing and fault diagnosis modules. 4. Fault Diagnosis The system uses continuous wavelet transform to process the acquired signals and performs fault diagnosis based on the transform results. The following example uses the acquired radial swing signal in the Y-axis direction of the turbine's main shaft (Figure 4) to illustrate the results of fault diagnosis using wavelet transform. A biorthogonal spline wavelet with a message moment of 4 is selected, and continuous wavelet transform is performed on the time-domain signal at a scale of [1:1:48]. Figure 5 shows the three-dimensional graphical representation of the wavelet coefficients of the swing signal at scale-time. It is evident from the figure that four periodically occurring modulus maxima curves appear, with the third curve always having a smaller slope. Under normal circumstances, a smooth cone would appear in the three-dimensional graph, indicating strong singularity at these points. This is indeed caused by sliding friction of the turbine's main shaft. The modulus maxima curves are obtained by setting a threshold algorithm, as shown in Figure 6. The intersection point of two modulus maxima curves can accurately pinpoint the fault location. This paper, considering the practical needs of hydro-generator condition monitoring, researches and develops a hydro-generator condition monitoring system based on LONWORKS fieldbus technology. The hardware and software development of the field data acquisition nodes was completed, and the system's data communication message protocol and communication software were designed. The system's host computer software uses an MS-SQLSERVER2000 database for data storage and, through wavelet transform, initially implements fault diagnosis.