Multi-protocol adaptive multi-functional data acquisition system and method based on FPGA technology

Abstract

The application relates to a multi-protocol adaptive multifunctional data acquisition system and method based on FPGA technology, and particularly relates to the field of data acquisition. The system matches the electrical and timing characteristics of original electrical signals of a physical channel with a pre-stored protocol feature library on an FPGA chip to identify the protocol type and load corresponding analysis logic by analyzing the original electrical signals of the physical channel in real time. Single sensor data sequences are subjected to numerical difference processing, the change rate and acceleration of physical quantities are calculated, and data point mutation probability is obtained by inputting a Bayesian network model built in the FPGA. A risk state vector is formed by combining the mutation probability of each sensor channel and the spatial position coordinates, Hadamard product operation is performed on the risk state vector and a preset disaster mode weight vector to obtain channel weight coefficients, and the data acquisition frequency and processing resource allocation are adjusted. Multi-channel data aligned with timestamps are subjected to tensor contraction operation with a pre-stored multi-dimensional failure mode feature kernel, and a fusion early warning data packet is generated and preferentially sent when the result exceeds a corresponding risk threshold.

Description

Technical Field

[0001] This application belongs to the field of data acquisition, and in particular relates to a multi-protocol adaptive multi-functional data acquisition system and method based on FPGA technology. Background Technology

[0002] Slope stability is directly related to the safety of people's lives and property and major engineering facilities in the surrounding area. To prevent and mitigate geological disasters such as landslides and collapses, establishing an efficient and reliable automated monitoring and early warning system is crucial. Traditional slope monitoring systems typically employ a sensor + data acquisition unit (DAU) + central server architecture. Front-end sensors, such as displacement gauges, rain gauges, pore water pressure gauges, and stress gauges, are responsible for sensing various physical quantities of the slope. These sensors are diverse, and their output signal data formats and communication protocols (e.g., RS485, CAN, SPI, I2C) vary. Existing data acquisition units usually use fixed hardware interfaces and software protocol stacks, resulting in poor compatibility and limited scalability. When it is necessary to connect new sensors or replace devices with devices using different protocols, it often requires hardware replacement or complex software secondary development, leading to high system integration and maintenance costs and difficulty in flexibly adapting to changing on-site monitoring needs. Furthermore, the data is only simply collected at the front end, while detailed analysis and decision-making are completed on a remote central server. The long data transmission link inevitably results in communication delays, which may lead to delayed early warning information and missed precious emergency response time for sudden landslide disasters.

[0003] Regarding data processing and early warning models, on the one hand, the data acquisition strategies of monitoring systems are mostly statically configured, meaning that all sensor channels are polled at a fixed sampling frequency, making it impossible to dynamically adjust the allocation of monitoring resources based on real-time changes in the risk status of different areas of the slope. For long-term stable areas, fixed high-frequency acquisition generates a large amount of redundant data, wasting storage and transmission resources; while for critical areas where risks are accumulating, a fixed acquisition frequency may be insufficient to capture the transient characteristics of impending disasters. On the other hand, existing early warning criteria are mostly based on empirical thresholds for single physical quantities, such as triggering an alarm only when the displacement rate or cumulative displacement at a certain measuring point exceeds a preset safety value. This method ignores the fact that slope instability is a complex process in which multiple physical quantities are coupled and evolve collaboratively. For example, the increase in pore water pressure caused by heavy rainfall leads to a decrease in soil shear strength and ultimately accelerates displacement, which is a typical multi-factor coupling chain. Single threshold models cannot effectively characterize this multi-dimensional correlation and disaster mode, resulting in low accuracy in early warning, prone to false alarms or missed alarms, and difficult to accurately identify and predict potential instability modes. Summary of the Invention

[0004] To address the above problems, this invention provides a multi-protocol adaptive multi-functional data acquisition system based on FPGA technology, comprising:

[0005] The identification module is used to acquire the raw electrical signals on the physical channel of the data acquisition access FPGA. By analyzing the electrical and timing characteristics of the raw electrical signals in real time, it matches them with the protocol feature library pre-set in the on-chip memory of the FPGA to identify the protocol type. The protocol feature library records the electrical characteristics, encoding format and frame structure template of various protocols, and loads the corresponding protocol parsing logic for the physical channel based on the identification results.

[0006] The mutation calculation module is used to process the single sensor data sequence obtained through the protocol parsing logic using the numerical difference method to calculate the rate of change and acceleration of the monitored physical quantity, and to calculate the mutation probability of the data point by taking the rate of change and acceleration as input through the Bayesian network model pre-installed in the FPGA.

[0007] The weight adjustment module is used to combine the data mutation probability of all sensor channels with their respective spatial deployment location coordinates to obtain a risk state vector. The risk state vector is then used to perform a Hadamard product operation with a preset disaster mode weight vector to obtain the weight coefficient of each sensor channel. Based on the weight coefficient, the data acquisition frequency and processing resource allocation of each sensor channel are adjusted.

[0008] The sending module is used to perform tensor shrinking operations on data from different sensor channels with timestamp alignment and multiple preset multidimensional failure mode feature kernels. The multidimensional failure mode feature kernels represent the coupling response relationship between multiple physical quantities under a specific slope instability mode. When the result of any tensor shrinking operation exceeds the corresponding risk trigger threshold, a fusion early warning data packet is generated and sent first.

[0009] Optionally, the step of identifying the protocol type by analyzing the electrical and timing characteristics of the original electrical signal in real time and matching it with a pre-built protocol feature library in the FPGA's on-chip memory includes:

[0010] The initial data stream on the physical channel is acquired, the high and low level voltages of the initial data stream are measured, and the data stream is compared with various protocol electrical characteristics recorded in the protocol feature library to obtain candidate protocols.

[0011] Within the candidate protocol range, the data transmission baud rate is estimated by measuring the shortest duration of consecutive identical voltage levels.

[0012] Based on the estimated baud rate, the data stream is decoded using multiple frame structure templates corresponding to the candidate protocols. The frame structure templates define a combination of start bits, data bits, parity bits, and stop bits. The first template that can successfully parse a continuous and valid data frame is taken as the protocol type of the current channel.

[0013] Optionally, the process of using the numerical difference method to calculate the rate of change and acceleration of the monitored physical quantity includes:

[0014] For the sensor reading S(t) at time t and the readings S(t-Δt) and S(t-2Δt) at previous times, where Δt is the sampling time interval;

[0015] The rate of change v(t) is calculated using the first-order backward difference formula: v(t) = [S(t) - S(t - Δt)] / Δt.

[0016] The acceleration a(t) is calculated using the second-order backward difference formula: a(t) = [S(t) - 2S(t - Δt) + S(t - 2Δt)] / (Δt) 2 .

[0017] Optionally, the step of performing a Hadamard product operation between the risk state vector and a preset disaster mode weight vector to obtain the weight coefficients of each sensor channel includes:

[0018] Load K preset disaster mode weight vectors {W1, W2, ..., W...} from the FPGA on-chip memory. k}, each N×M dimensional matrix W k It corresponds to a specific disaster mode;

[0019] Connect the risk state vector R with each W. k Performing the Hadamard product operation yields K intermediate weight matrices D. k =R⊙W k ;

[0020] A weight matrix D is generated by taking the maximum value of each of the K intermediate weight matrices. Any element in D is a weight matrix D. ij =max(D 1ij D 2ij ,...,D kij ).

[0021] Optionally, adjusting the data acquisition frequency and processing resource allocation for each sensor channel based on the weighting coefficient includes:

[0022] The dynamic weighting coefficient D of each sensor channel ij Mapped to a preset frequency range [F min ,F max Within [the specified range], determine the data acquisition frequency F.ij F ij It is D ij A monotonically increasing function;

[0023] The weighting coefficient D ij For channels that exceed the high-risk threshold, the data processing logic is allocated additional parallel processing units or higher bus access priority.

[0024] Optionally, the step of performing tensor shrinking operation on data from different sensor channels and aligned with timestamps with multiple preset multidimensional failure mode feature kernels includes:

[0025] At the same timestamp t, with a certain sensor as the center, extract the data of K different physical quantities measured by all sensors in the P×Q spatial neighborhood to obtain the third-order data tensor T of P×Q×K.

[0026] Multiple pre-set third-order failure mode feature kernels {K1, K2, ...} are loaded from the FPGA on-chip memory. Each feature kernel K m The dimension is P×Q×K, and the element value represents the ideal coupling response strength between multiple physical quantities in the neighborhood under a specific instability mode;

[0027] Connect the data tensor T with each feature kernel K m Element-wise product summation, or tensor shrinking, is performed to obtain the matching risk value R of the neighborhood with each failure mode at time t. m =Σ i Σ j Σ k (T ijk ×K mijk ).

[0028] This invention also provides a multi-protocol adaptive multi-functional data acquisition method based on FPGA technology, including:

[0029] The system acquires the raw electrical signals on the physical channel of the FPGA data acquisition system, analyzes the electrical and timing characteristics of the raw electrical signals in real time, and matches them with the protocol feature library pre-set in the FPGA on-chip memory to identify the protocol type. The protocol feature library records the electrical characteristics, encoding format and frame structure template of various protocols, and loads the corresponding protocol parsing logic for the physical channel based on the identification results.

[0030] For a single sensor data sequence obtained through the protocol parsing logic, the numerical difference method is used to process it, calculate the rate of change and acceleration of the monitored physical quantity, and use the rate of change and acceleration as input to calculate the abrupt change probability of the data point through a Bayesian network model pre-installed in the FPGA.

[0031] The risk state vector is obtained by combining the data mutation probability of all sensor channels with their corresponding spatial deployment location coordinates. The risk state vector is then used to perform a Hadamard product operation with a preset disaster mode weight vector to obtain the weight coefficient of each sensor channel. The data acquisition frequency and processing resource allocation of each sensor channel are adjusted according to the weight coefficient.

[0032] Data from different sensor channels with timestamp alignment are subjected to tensor shrinkage operations with multiple preset multidimensional failure mode feature kernels. The multidimensional failure mode feature kernels characterize the coupling response relationship between multiple physical quantities under a specific slope instability mode. When the result of any tensor shrinkage operation exceeds the corresponding risk trigger threshold, a fusion early warning data packet is generated and sent first.

[0033] Optionally, the step of identifying the protocol type by analyzing the electrical and timing characteristics of the original electrical signal in real time and matching it with a pre-built protocol feature library in the FPGA's on-chip memory includes:

[0034] The initial data stream on the physical channel is acquired, the high and low level voltages of the initial data stream are measured, and the data stream is compared with various protocol electrical characteristics recorded in the protocol feature library to obtain candidate protocols.

[0035] Within the candidate protocol range, the data transmission baud rate is estimated by measuring the shortest duration of consecutive identical voltage levels.

[0036] Based on the estimated baud rate, the data stream is decoded using multiple frame structure templates corresponding to the candidate protocols. The frame structure templates define a combination of start bits, data bits, parity bits, and stop bits. The first template that can successfully parse a continuous and valid data frame is taken as the protocol type of the current channel.

[0037] Optionally, the process of using the numerical difference method to calculate the rate of change and acceleration of the monitored physical quantity includes:

[0038] For the sensor reading S(t) at time t and the readings S(t-Δt) and S(t-2Δt) at previous times, where Δt is the sampling time interval;

[0039] The rate of change v(t) is calculated using the first-order backward difference formula: v(t) = [S(t) - S(t - Δt)] / Δt.

[0040] The acceleration a(t) is calculated using the second-order backward difference formula: a(t) = [S(t) - 2S(t - Δt) + S(t - 2Δt)] / (Δt) 2 .

[0041] Optionally, the step of performing a Hadamard product operation between the risk state vector and a preset disaster mode weight vector to obtain the weight coefficients of each sensor channel includes:

[0042] Load K preset disaster mode weight vectors {W1, W2, ..., W...} from the FPGA on-chip memory. k}, each N×M dimensional matrix W k It corresponds to a specific disaster mode;

[0043] Connect the risk state vector R with each W. k Performing the Hadamard product operation yields K intermediate weight matrices D. k =R⊙W k ;

[0044] A weight matrix D is generated by taking the maximum value of each of the K intermediate weight matrices. Any element in D is a weight matrix D. ij =max(D 1ij D 2ij ,...,D kij ).

[0045] Optionally, adjusting the data acquisition frequency and processing resource allocation for each sensor channel based on the weighting coefficient includes:

[0046] The dynamic weighting coefficient D of each sensor channel ij Mapped to a preset frequency range [F min ,F max Within [the specified range], determine the data acquisition frequency F. ij F ij It is D ij A monotonically increasing function;

[0047] The weighting coefficient D ij For channels that exceed the high-risk threshold, the data processing logic is allocated additional parallel processing units or higher bus access priority.

[0048] Optionally, the step of performing tensor shrinking operation on data from different sensor channels and aligned with timestamps with multiple preset multidimensional failure mode feature kernels includes:

[0049] At the same timestamp t, with a certain sensor as the center, extract the data of K different physical quantities measured by all sensors in the P×Q spatial neighborhood to obtain the third-order data tensor T of P×Q×K.

[0050] Multiple pre-set third-order failure mode feature kernels {K1, K2, ...} are loaded from the FPGA on-chip memory. Each feature kernel K mThe dimension is P×Q×K, and the element value represents the ideal coupling response strength between multiple physical quantities in the neighborhood under a specific instability mode;

[0051] Connect the data tensor T with each feature kernel K m Element-wise product summation, or tensor shrinking, is performed to obtain the matching risk value R of the neighborhood with each failure mode at time t. m =Σ i Σ j Σ k (T ijk ×K mijk ).

[0052] This invention achieves unified access to sensors with various communication protocols by directly analyzing the electrical and temporal characteristics of raw electrical signals at the hardware level, enhancing system compatibility and scalability while reducing system integration complexity and cost. Furthermore, by introducing a Bayesian network model at the data acquisition front end, probabilistic evaluation of the changing trends of data from individual measurement points enables the earlier detection of potential anomalies compared to traditional threshold methods, improving early warning sensitivity. Moreover, this invention utilizes information from all measurement points to construct a risk state vector and combines it with a multi-dimensional failure mode feature kernel for tensor operations, uncovering the coupling correlations between multiple physical quantities during slope instability. This allows early warning decisions to be based on accurate identification of the overall disaster mode, thereby improving the accuracy and reliability of early warnings and reducing the risk of false alarms and missed alarms. Attached Figure Description

[0053] Figure 1 This is a structural diagram of the protocol identification in an embodiment of the present invention;

[0054] Figure 2 This is a schematic diagram illustrating the data mutation probability calculation process in an embodiment of the present invention;

[0055] Figure 3 This is a schematic diagram illustrating the generation of weighting coefficients in an embodiment of the present invention;

[0056] Figure 4 This is a schematic diagram of tensor contraction based on an embodiment of the present invention. Detailed Implementation

[0057] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0058] The terms "first," "second," and corresponding designations in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of units is not necessarily limited to those units but may include other units not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0059] Furthermore, in the description of this application, unless otherwise stated, "multiple" means two or more. The terms "and / or" or the character " / " in this application are merely descriptions of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B, or A / B, can represent: A existing alone, A and B existing simultaneously, and B existing alone.

[0060] In a specific embodiment, the present invention provides a multi-protocol adaptive multi-functional data acquisition system based on FPGA technology, comprising:

[0061] The identification module is used to acquire the raw electrical signals on the physical channel of the data acquisition access FPGA. By analyzing the electrical and timing characteristics of the raw electrical signals in real time, it matches them with the protocol feature library pre-set in the on-chip memory of the FPGA to identify the protocol type. The protocol feature library records the electrical characteristics, encoding format and frame structure template of various protocols, and loads the corresponding protocol parsing logic for the physical channel based on the identification results.

[0062] The FPGA's SERDES transceiver unit directly samples the voltage waveform of the physical channel at high frequency, using a digital signal processing module to analyze its high and low level durations, edge intervals, and idle state levels. For example, if a differential voltage is detected and the idle state is high, and the bit width is measured to be approximately 104 microseconds, then the matching is an RS485 protocol with a baud rate of 9600. Figure 1 As shown, the protocol feature library is a lookup table stored in the FPGA's internal block RAM, recording information such as the differential levels, bit timing parameters, and frame ID structure of the CAN bus, or the SCL and SDA timing relationships of the I2C bus. Upon successful matching, a hardware description language module corresponding to the protocol, such as a dedicated VHDL module for parsing Modbus-RTU frames, stored in external flash memory, is loaded into the FPGA's programmable logic array to serve that channel, using partial reconstruction techniques.

[0063] The mutation calculation module is used to process the single sensor data sequence obtained through the protocol parsing logic using the numerical difference method to calculate the rate of change and acceleration of the monitored physical quantity, and to calculate the mutation probability of the data point by taking the rate of change and acceleration as input through a Bayesian network model pre-installed in the FPGA.

[0064] Specifically, for a displacement sensor acquiring a value X at time t t In one embodiment, its rate of change, V, is calculated using the backward difference formula. t For X t Subtract X t-1 Then, divide by the sampling time interval; the changing acceleration is A. t For V t Subtract V t-1 Then, divide by the sampling time interval. The Bayesian network model is implemented as a hardware inference circuit within an FPGA. The calculated velocity and acceleration values ​​are discretized into low, medium, and high levels, then input into the Bayesian network model. The model is then computed in parallel in hardware using a belief propagation algorithm, outputting the posterior probability value representing a sudden change at that data point, such as... Figure 2 As shown. In another embodiment, filtering is performed before calculating the mutation probability to reduce the impact of noise.

[0065] The weight adjustment module is used to combine the data mutation probability of all sensor channels with their respective spatial deployment location coordinates to obtain a risk state vector. The risk state vector is then used to perform a Hadamard product operation with a preset disaster mode weight vector to obtain the weight coefficient of each sensor channel. Based on the weight coefficient, the data acquisition frequency and processing resource allocation of each sensor channel are adjusted.

[0066] If N sensors are deployed on the slope, the mutation probabilities P1 to PN of all channels constitute an N-dimensional risk state vector. A pre-defined catastrophic mode weight vector, such as for a slope crest traction landslide mode, will assign higher initial weight values ​​to the sensors deployed at the slope crest. The Hadamard product operation involves element-wise multiplication of these two N-dimensional vectors to obtain a new N-dimensional dynamic weight vector, such as... Figure 3 As shown. For example, if the mutation probability of the i-th sensor is 0.6 and its correlation weight in this catastrophe mode is 0.8, then its dynamic weight coefficient is 0.48. Based on the above coefficient value, the scheduler inside the FPGA changes the sampling clock frequency of the i-th sensor channel by modifying the configuration register of a programmable clock divider, and at the same time allocates more pipeline processing units to it by controlling the gating clock of the data path.

[0067] The sending module is used to perform tensor shrinking operations on data from different sensor channels with timestamp alignment and multiple preset multidimensional failure mode feature kernels. The multidimensional failure mode feature kernels represent the coupling response relationship between multiple physical quantities under a specific slope instability mode. When the result of any tensor shrinking operation exceeds the corresponding risk trigger threshold, a fusion early warning data packet is generated and sent first.

[0068] The FPGA appends a high-precision hardware timestamp to each acquired data point. For example, three data values ​​from a displacement gauge, pore water pressure gauge, and rain gauge within the same time window are combined into a third-order data tensor. The multidimensional failure mode feature kernel is a high-order tensor pre-loaded in on-chip memory, such as a feature kernel characterizing rainfall-induced infiltration landslides. Its element values ​​define the strong coupling relationship between the three physical quantities: heavy rainfall, a sharp rise in pore water pressure, and displacement acceleration. Tensor shrinking operations are implemented in the FPGA using a dedicated multiply-accumulate array, multiplying and adding the data tensor and the feature kernel tensor along the corresponding dimensions to obtain a scalar result. The magnitude of this scalar characterizes the degree of matching between the current multi-physical quantity state and the specific failure mode. If the result is greater than the risk threshold set for that mode, such as 0.9, a warning message containing the warning level, trigger mode ID, relevant sensor data, and timestamp is immediately encapsulated and sent through a priority queue of the communication interface. During timestamp alignment, interpolation is performed if the data lengths are different.

[0069] In an optional embodiment, the step of identifying the protocol type by analyzing the electrical and timing characteristics of the original electrical signal in real time and matching it with a pre-built protocol feature library in the FPGA's on-chip memory includes:

[0070] The initial data stream on the physical channel is acquired, the high and low level voltages of the initial data stream are measured, and the data stream is compared with various protocol electrical characteristics recorded in the protocol feature library to obtain candidate protocols.

[0071] Within the candidate protocol range, the data transmission baud rate is estimated by measuring the shortest duration of consecutive identical voltage levels.

[0072] Based on the estimated baud rate, the data stream is decoded using multiple frame structure templates corresponding to the candidate protocols. The frame structure templates define a combination of start bits, data bits, parity bits, and stop bits. The first template that can successfully parse a continuous and valid data frame is taken as the protocol type of the current channel.

[0073] When an unknown serial device is connected to the system, the voltage on its signal lines is measured. If the high level is measured to be -5V and the low level to be +5V, it is preliminarily determined that it is likely an RS-232 protocol, thus ruling out protocols such as CAN and I2C. The shortest pulse width appearing in the signal stream is measured, for example, approximately 104 microseconds, from which the baud rate is estimated to be close to 9600 bits per second. Using a baud rate of 9600, and applying a frame structure template of 8 data bits, no parity bit, and 1 stop bit, decoding is attempted. If multiple data frames can be successfully decoded consecutively, the protocol of the channel is finally confirmed to be RS-232 and the baud rate is 9600.

[0074] In an optional embodiment, the step of using the numerical difference method to calculate the rate of change and acceleration of the monitored physical quantity includes:

[0075] For the sensor reading S(t) at time t and the readings S(t-Δt) and S(t-2Δt) at previous times, where Δt is the sampling time interval;

[0076] The rate of change v(t) is calculated using the first-order backward difference formula: v(t) = [S(t) - S(t - Δt)] / Δt.

[0077] The acceleration a(t) is calculated using the second-order backward difference formula: a(t) = [S(t) - 2S(t - Δt) + S(t - 2Δt)] / (Δt) 2 .

[0078] Assuming a temperature sensor has a sampling interval Δt of 1 second, at the current time t, the measured temperature S(t) is 30 degrees Celsius. The previous second, S(t-Δt) was 25 degrees Celsius, and the second before that, S(t-2Δt) was 24 degrees Celsius. Based on this, the rate of temperature change v(t) is calculated to be 5 degrees Celsius per second, and the acceleration a(t) is calculated to be 4 degrees Celsius per square second. These two calculated rate values ​​of 5 and acceleration values ​​of 4 are input as new evidence into the Bayesian network to determine whether a sudden change in temperature has occurred.

[0079] In an optional embodiment, the step of performing a Hadamard product operation using the risk state vector and a preset disaster mode weight vector to obtain the weight coefficients of each sensor channel includes:

[0080] Load K preset disaster mode weight vectors {W1, W2, ..., W...} from the FPGA on-chip memory. k}, each N×M dimensional matrix W k It corresponds to a specific disaster mode;

[0081] Connect the risk state vector R with each W. k Performing the Hadamard product operation yields K intermediate weight matrices D. k=R⊙W k ;

[0082] A weight matrix D is generated by taking the maximum value of each of the K intermediate weight matrices. Any element in D is a weight matrix D. ij =max(D 1ij D 2ij ,...,D kij ).

[0083] Consider a sensor located in row i, column j, with a current risk value of 0.8. Two disaster modes are pre-stored: fire and flood. The weight of this sensor's location in the fire mode is 0.9, and its weight in the flood mode is 0.1. The associated risk of this sensor under each mode is calculated, resulting in a value of 0.08. The maximum value of these two values, 0.72, is taken as the final dynamic weight coefficient for this sensor. This process is performed once for each sensor in the matrix to generate a complete dynamic weight matrix.

[0084] In an optional embodiment, adjusting the data acquisition frequency and processing resource allocation for each sensor channel based on the weighting coefficient includes:

[0085] The dynamic weighting coefficient D of each sensor channel ij Mapped to a preset frequency range [F min ,F max Within [the specified range], determine the data acquisition frequency F. ij F ij It is D ij A monotonically increasing function;

[0086] The weighting coefficient D ij For channels that exceed the high-risk threshold, the data processing logic is allocated additional parallel processing units or higher bus access priority.

[0087] Specifically, assuming the dynamic weighting coefficient ranges from 0 to 1, and the preset data acquisition frequency range is 10 Hz to 1000 Hz, a low-risk sensor channel with a weighting coefficient of 0.1 might have its acquisition frequency set to a lower 50 Hz. When the weighting coefficient of another sensor channel spikes to 0.9 due to anomalies, its acquisition frequency will be dynamically increased to 901 Hz through linear mapping. Simultaneously, since this weight of 0.9 exceeds the preset high-risk threshold of 0.8, a dedicated digital signal processing unit will be immediately allocated to it from the FPGA resource pool, and its access priority on the data bus will be increased to ensure that critical data is processed in the most timely manner.

[0088] In an optional embodiment, the step of performing tensor shrinking operation on data from different sensor channels and aligned with timestamps with multiple preset multidimensional failure mode feature kernels includes:

[0089] At the same timestamp t, with a certain sensor as the center, extract the data of K different physical quantities measured by all sensors in the P×Q spatial neighborhood to obtain the third-order data tensor T of P×Q×K.

[0090] Multiple pre-set third-order failure mode feature kernels {K1, K2, ...} are loaded from the FPGA on-chip memory. Each feature kernel K m The dimension is P×Q×K, and the element value represents the ideal coupling response strength between multiple physical quantities in the neighborhood under a specific instability mode;

[0091] Connect the data tensor T with each feature kernel K m Element-wise product summation, or tensor shrinking, is performed to obtain the matching risk value R of the neighborhood with each failure mode at time t. m =Σ i Σ j Σ k (T ijk ×K mijk ).

[0092] Consider monitoring a key sensor or location. Using the key sensor as the center, extract temperature and vibration data within its 3x3 spatial neighborhood, forming a 3x3x2 data tensor T. Simultaneously, load a pre-defined failure mode feature kernel K1 from memory. This kernel's numerical distribution is characterized by a high temperature weight in the central region and a relatively high vibration weight in both the central and adjacent regions. By multiplying each of the 18 elements of the data tensor T by the corresponding 18 elements of the feature kernel K1 and then summing them all, a scalar risk value is obtained, for example, 0.92. A higher value indicates a closer match between the current state and the pre-defined failure mode. Figure 4 As shown.

[0093] In a second specific embodiment, the present invention also provides a multi-protocol adaptive multi-functional data acquisition method based on FPGA technology, including:

[0094] The system acquires the raw electrical signals on the physical channel of the FPGA data acquisition system, analyzes the electrical and timing characteristics of the raw electrical signals in real time, and matches them with the protocol feature library pre-set in the FPGA on-chip memory to identify the protocol type. The protocol feature library records the electrical characteristics, encoding format and frame structure template of various protocols, and loads the corresponding protocol parsing logic for the physical channel based on the identification results.

[0095] For a single sensor data sequence obtained through the protocol parsing logic, the numerical difference method is used to process it, calculate the rate of change and acceleration of the monitored physical quantity, and use the rate of change and acceleration as input to calculate the abrupt change probability of the data point through a Bayesian network model pre-installed in the FPGA.

[0096] The risk state vector is obtained by combining the data mutation probability of all sensor channels with their corresponding spatial deployment location coordinates. The risk state vector is then used to perform a Hadamard product operation with a preset disaster mode weight vector to obtain the weight coefficient of each sensor channel. The data acquisition frequency and processing resource allocation of each sensor channel are adjusted according to the weight coefficient.

[0097] Data from different sensor channels, aligned with time stamps, are subjected to tensor shrinkage operations with multiple pre-set multidimensional failure mode feature kernels. These multidimensional failure mode feature kernels characterize the coupling response relationships between multiple physical quantities under specific slope instability modes. When the result of any tensor shrinkage operation exceeds the corresponding risk trigger threshold, a fused early warning data packet is generated and sent preferentially. In one specific embodiment, the step of identifying the protocol type by real-time analysis of the electrical and timing characteristics of the original electrical signal and matching it with a pre-set protocol feature library in the FPGA's on-chip memory includes:

[0098] The initial data stream on the physical channel is acquired, the high and low level voltages of the initial data stream are measured, and the data stream is compared with various protocol electrical characteristics recorded in the protocol feature library to obtain candidate protocols.

[0099] Within the candidate protocol range, the data transmission baud rate is estimated by measuring the shortest duration of consecutive identical voltage levels.

[0100] Based on the estimated baud rate, the data stream is decoded using multiple frame structure templates corresponding to the candidate protocols. The frame structure templates define a combination of start bits, data bits, parity bits, and stop bits. The first template that can successfully parse a continuous and valid data frame is taken as the protocol type of the current channel.

[0101] In one specific embodiment, the step of using the numerical difference method to calculate the rate of change and acceleration of the monitored physical quantity includes:

[0102] For the sensor reading S(t) at time t and the readings S(t-Δt) and S(t-2Δt) at previous times, where Δt is the sampling time interval;

[0103] The rate of change v(t) is calculated using the first-order backward difference formula: v(t) = [S(t) - S(t - Δt)] / Δt.

[0104] The acceleration a(t) is calculated using the second-order backward difference formula: a(t) = [S(t) - 2S(t - Δt) + S(t - 2Δt)] / (Δt) 2 .

[0105] In one specific embodiment, the step of performing a Hadamard product operation using the risk state vector and a preset disaster mode weight vector to obtain the weight coefficients of each sensor channel includes:

[0106] Load K preset disaster mode weight vectors {W1, W2, ..., W...} from the FPGA on-chip memory. k}, each N×M dimensional matrix W k It corresponds to a specific disaster mode;

[0107] Connect the risk state vector R with each W. k Performing the Hadamard product operation yields K intermediate weight matrices D. k =R⊙W k ;

[0108] A weight matrix D is generated by taking the maximum value of each of the K intermediate weight matrices. Any element in D is a weight matrix D. ij =max(D 1ij D 2ij ,...,D kij ).

[0109] In one specific embodiment, adjusting the data acquisition frequency and processing resource allocation for each sensor channel based on the weighting coefficient includes:

[0110] The dynamic weighting coefficient D of each sensor channel ij Mapped to a preset frequency range [F min ,F max Within [the specified range], determine the data acquisition frequency F. ij F ij It is D ij A monotonically increasing function;

[0111] The weighting coefficient D ij For channels that exceed the high-risk threshold, the data processing logic is allocated additional parallel processing units or higher bus access priority.

[0112] In one specific embodiment, the step of performing tensor shrinking operation on data from different sensor channels and aligned with timestamps with multiple preset multidimensional failure mode feature kernels includes:

[0113] At the same timestamp t, with a certain sensor as the center, extract the data of K different physical quantities measured by all sensors in the P×Q spatial neighborhood to obtain the third-order data tensor T of P×Q×K.

[0114] Multiple pre-set third-order failure mode feature kernels {K1, K2, ...} are loaded from the FPGA on-chip memory. Each feature kernel K mThe dimension is P×Q×K, and the element value represents the ideal coupling response strength between multiple physical quantities in the neighborhood under a specific instability mode;

[0115] Connect the data tensor T with each feature kernel K m Element-wise product summation, or tensor shrinking, is performed to obtain the matching risk value R of the neighborhood with each failure mode at time t. m =Σ i Σ j Σ k (T ijk ×K mijk ).

[0116] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some feature data may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units through some interfaces, and may be electrical, mechanical, or other forms.

[0117] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0118] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0119] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A multi-protocol adaptive multi-functional data acquisition system based on FPGA technology, characterized in that, include: The identification module is used to acquire the raw electrical signals on the physical channel of the data acquisition access FPGA. By analyzing the electrical and timing characteristics of the raw electrical signals in real time, it matches them with the protocol feature library pre-set in the on-chip memory of the FPGA to identify the protocol type. The protocol feature library records the electrical characteristics, encoding format and frame structure template of various protocols, and loads the corresponding protocol parsing logic for the physical channel based on the identification results. The mutation calculation module is used to process the single sensor data sequence obtained through the protocol parsing logic using the numerical difference method to calculate the rate of change and acceleration of the monitored physical quantity, and to calculate the mutation probability of the data point by taking the rate of change and acceleration as input through the Bayesian network model pre-installed in the FPGA. The weight adjustment module is used to combine the mutation probability of all sensor channels with their corresponding spatial deployment location coordinates to obtain a risk state vector. The risk state vector is then used to perform a Hadamard product operation with a preset disaster mode weight vector to obtain the weight coefficient of each sensor channel. Based on the weight coefficient, the data acquisition frequency and processing resource allocation of each sensor channel are adjusted. The sending module is used to perform tensor shrinking operations on data from different sensor channels with timestamp alignment and multiple preset multidimensional failure mode feature kernels. The multidimensional failure mode feature kernels represent the coupling response relationship between multiple physical quantities under a specific slope instability mode. When the result of any tensor shrinking operation exceeds the corresponding risk trigger threshold, a fusion early warning data packet is generated and sent first.

2. The system according to claim 1, characterized in that, The process of identifying the protocol type by analyzing the electrical and timing characteristics of the original electrical signal in real time and matching it with a pre-built protocol feature library in the FPGA's on-chip memory includes: The initial data stream on the physical channel is acquired, the high and low level voltages of the initial data stream are measured, and the data stream is compared with various protocol electrical characteristics recorded in the protocol feature library to obtain candidate protocols. Within the candidate protocol range, the data transmission baud rate is estimated by measuring the shortest duration of consecutive identical voltage levels. Based on the estimated data transmission baud rate, the data stream is decoded using multiple frame structure templates corresponding to candidate protocols. The frame structure templates define a combination of start bits, data bits, parity bits, and stop bits. The first template that can successfully parse a continuous and valid data frame is taken as the protocol type of the current channel.

3. The system according to claim 1, characterized in that, The numerical difference method is used to process and calculate the rate of change and acceleration of the monitored physical quantity, including: For the sensor reading S(t) at time t and the readings S(t-Δt) and S(t-2Δt) at previous times, where Δt is the sampling time interval; The rate of change v(t) = [S(t) - S(t - Δt)] / Δt is calculated using the first-order backward difference formula; The acceleration a(t) is calculated using the second-order backward difference formula: a(t) = [S(t) - 2S(t-Δt) + S(t-2Δt)] / (Δt)².

4. The system according to claim 1, characterized in that, The step of performing a Hadamard product operation using the risk state vector and a preset disaster mode weight vector to obtain the weight coefficients of each sensor channel includes: Load K preset disaster mode weight vectors from the FPGA on-chip memory. Each N×M dimensional matrix It corresponds to a specific disaster mode; The risk state vector R is respectively connected to each Performing the Hadamard product operation yields K intermediate weight matrices. ; A weight matrix D is generated by taking the maximum value of each of the K intermediate weight matrices. Any element in D... .

5. The system according to claim 1, characterized in that, The adjustment of the data acquisition frequency and processing resource allocation for each sensor channel based on the weighting coefficient includes: Dynamic weighting coefficients for each sensor channel Mapped to a preset frequency range [F min ,F max Within [the specified area], determine the data acquisition frequency. ,in yes A monotonically increasing function; Weighting coefficients For channels that exceed the high-risk threshold, the data processing logic is allocated additional parallel processing units or higher bus access priority.

6. The system according to claim 1, characterized in that, The step of performing tensor shrinking operations on data from different sensor channels that are time-stamp aligned with multiple preset multidimensional failure mode feature kernels includes: At the same timestamp t, with a certain sensor as the center, extract the data of K different physical quantities measured by all sensors in the P×Q spatial neighborhood to obtain the third-order data tensor T of P×Q×K. Load multiple pre-defined third-order failure mode signatures from the FPGA on-chip memory. Each feature kernel The dimension is P×Q×K, and the element value represents the ideal coupling response strength between multiple physical quantities in the neighborhood under a specific instability mode; Connect the data tensor T with each feature kernel Element-wise product summation, or tensor shrinking, is performed to obtain the matching risk value of the neighborhood with each failure mode at time t. .

7. A multi-protocol adaptive multi-functional data acquisition method based on FPGA technology, characterized in that, include: The system acquires the raw electrical signals on the physical channel of the FPGA data acquisition system, analyzes the electrical and timing characteristics of the raw electrical signals in real time, and matches them with the protocol feature library pre-set in the FPGA on-chip memory to identify the protocol type. The protocol feature library records the electrical characteristics, encoding format and frame structure template of various protocols, and loads the corresponding protocol parsing logic for the physical channel based on the identification results. For a single sensor data sequence obtained through the protocol parsing logic, the numerical difference method is used to process it, calculate the rate of change and acceleration of the monitored physical quantity, and use the rate of change and acceleration as input to calculate the abrupt change probability of the data point through a Bayesian network model pre-installed in the FPGA. The mutation probability of all sensor channels is combined with their corresponding spatial deployment location coordinates to obtain a risk state vector. The risk state vector is then used to perform a Hadamard product operation with a preset disaster mode weight vector to obtain the weight coefficient of each sensor channel. The data acquisition frequency and processing resource allocation of each sensor channel are adjusted according to the weight coefficient. Data from different sensor channels with timestamp alignment are subjected to tensor shrinkage operations with multiple preset multidimensional failure mode feature kernels. The multidimensional failure mode feature kernels characterize the coupling response relationship between multiple physical quantities under a specific slope instability mode. When the result of any tensor shrinkage operation exceeds the corresponding risk trigger threshold, a fusion early warning data packet is generated and sent first.

8. The method according to claim 7, characterized in that, The process of identifying the protocol type by analyzing the electrical and timing characteristics of the original electrical signal in real time and matching it with a pre-built protocol feature library in the FPGA's on-chip memory includes: The initial data stream on the physical channel is acquired, the high and low level voltages of the initial data stream are measured, and the data stream is compared with various protocol electrical characteristics recorded in the protocol feature library to obtain candidate protocols. Within the candidate protocol range, the data transmission baud rate is estimated by measuring the shortest duration of consecutive identical voltage levels. Based on the estimated data transmission baud rate, the data stream is decoded using multiple frame structure templates corresponding to candidate protocols. The frame structure templates define a combination of start bits, data bits, parity bits, and stop bits. The first template that can successfully parse a continuous and valid data frame is taken as the protocol type of the current channel.

9. The method according to claim 7, characterized in that, The numerical difference method is used to process and calculate the rate of change and acceleration of the monitored physical quantity, including: For the sensor reading S(t) at time t and the readings S(t-Δt) and S(t-2Δt) at previous times, where Δt is the sampling time interval; The rate of change v(t) = [S(t) - S(t - Δt)] / Δt is calculated using the first-order backward difference formula; The acceleration a(t) is calculated using the second-order backward difference formula: a(t) = [S(t) - 2S(t-Δt) + S(t-2Δt)] / (Δt)².

10. The method according to claim 7, characterized in that, The step of performing a Hadamard product operation using the risk state vector and a preset disaster mode weight vector to obtain the weight coefficients of each sensor channel includes: Load K preset disaster mode weight vectors from the FPGA on-chip memory. Each N×M dimensional matrix It corresponds to a specific disaster mode; The risk state vector R is respectively connected to each Performing the Hadamard product operation yields K intermediate weight matrices. ; A weight matrix D is generated by taking the maximum value of each of the K intermediate weight matrices. Any element in D... .

Images