Communication data detection method based on intelligent safety helmet

Abstract

The invention discloses a communication data detection method based on an intelligent safety helmet. The method comprises the following steps: S1, collecting multi-modal sensor data of the intelligent safety helmet; step S2, preprocessing the multi-modal sensor data to form a standardized data stream with a unified timestamp; s3, packaging the preprocessed standardized data stream into a data frame, wherein a frame header comprises a serial number, a frame type and a data priority identifier; s4, calculating a dynamic fusion check code for each data frame, wherein the dynamic fusion check code is calculated and obtained through a multivariable hash function based on the load data, the equipment unique identifier and the real-time channel state parameter; the communication accuracy and safety of the intelligent safety helmet can be improved, operation and maintenance prediction is realized, and data protection is enhanced.

Description

Technical Field

[0001] This invention relates to the field of detection technology, and specifically to a method for detecting communication data based on a smart safety helmet. Background Technology

[0002] Smart safety helmets are core equipment in high-risk industrial work scenarios (such as construction, mining, and chemical inspection) that combine safety protection with intelligent sensing and communication. They not only have the head impact protection function of traditional safety helmets, but also integrate multi-modal sensors, wireless communication modules (such as 4G / 5G, LoRa, Bluetooth, etc.) and local storage units. They can collect voice interaction data, body posture data (such as whether they have fallen or whether they have bent over in violation of regulations) and work environment data (such as temperature, humidity, and concentration of harmful gases) from workers in real time, and transmit these data to the monitoring center. At the same time, they can receive instructions from the center and are a key carrier for realizing collaborative monitoring and safety early warning at the work site.

[0003] However, in practical applications of smart safety helmets in complex industrial scenarios, existing communication data detection technologies have multiple defects, making it difficult to meet the requirements of data transmission accuracy, reliability, security, and operational efficiency. These problems include poor adaptability to multimodal data processing, insufficient reliability of data verification, imbalance between encoding transmission efficiency and reliability, lack of flexibility in error handling, and insufficient local data security protection. Therefore, a communication data detection method based on smart safety helmets is proposed. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a communication data detection method based on a smart safety helmet, comprising the following steps:

[0005] Step S1: Collect multimodal sensor data from the smart safety helmet. The multimodal sensor data includes at least audio data, posture data, and environmental data.

[0006] Step S2: Preprocess the multimodal sensor data, including data alignment, noise reduction and normalization, to form a standardized data stream with a unified timestamp;

[0007] Step S3: Encapsulate the preprocessed standardized data stream into data frames. Each data frame includes a frame header, payload data, and a frame trailer. The frame header includes a sequence number, frame type, and data priority identifier.

[0008] Step S4: Calculate a dynamic fusion check code for each data frame, wherein the dynamic fusion check code is obtained by calculating the payload data, the device unique identifier, and the real-time channel state parameters through a multivariate hash function;

[0009] Step S5: Encode and transmit the data frame using an adaptive coding strategy based on channel state;

[0010] Step S6: At the receiving end, the data frame is received, and the legality of the dynamic fusion check code is verified. If the verification fails, the hierarchical error handling mechanism is adaptively adjusted and executed.

[0011] Step S7: Perform predictive maintenance analysis based on historical logs detected by communication data;

[0012] Step S8: Visualize and retrospectively analyze the communication quality data.

[0013] Furthermore, in step S4, the calculation process of the dynamic fusion check code includes:

[0014] First, the payload data D of the current data frame to be transmitted i The smart helmet's unique hardware identifier (UUID) and channel status parameters (CSI) obtained in real time through the communication module. i As three basic input elements;

[0015] Next, adaptive weight coefficients α, β, and γ are assigned to these three input elements respectively;

[0016] Then, the weighted three element data are subjected to a bitwise XOR operation, with the bitwise XOR operator being ⊕, to achieve the initial fusion of the data and generate a composite feature value;

[0017] Finally, to ensure the unpredictability and collision resistance of this composite feature value, it is input into a preset hash function H for one-way hash calculation. This hash function uses the SHA-256 algorithm, and the mathematical expression of the calculation process is as follows:

[0018] ;

[0019] in, This represents the final generated dynamic fusion checksum of the i-th frame, which is closely related to the current communication context.

[0020] Furthermore, the adaptive weight coefficients α, β, and γ are not fixed values, but are dynamically adjusted after performing sliding window statistical analysis on historical communication success rates and data integrity indicators. The adjustment strategy is as follows: the higher the historical error rate of the data dimension, the larger the corresponding weight coefficient is assigned in the next period, so as to enhance the contribution of the verification dimension in the overall verification.

[0021] Furthermore, in step S5, the adaptive coding strategy based on channel state specifically includes:

[0022] First, the two key physical layer parameters, Received Signal Strength Indicator (RSSI) and Signal-to-Noise Ratio (SNR), are obtained in real time through the underlying driver of the communication module, and then linearly combined into a normalized comprehensive channel quality index, Q.

[0023] Then, a rate adjustment function is set. The core feature of this function is that when the channel quality is below a certain threshold, the coding rate can be gradually reduced to prioritize reliability, while when the channel quality is good, the coding rate can be quickly increased to pursue transmission efficiency.

[0024] This function uses an S-shaped curve to simulate this adaptive adjustment behavior, and its mathematical expression is used to accurately calculate the optimal coding rate to be selected under the current channel conditions. The expression is as follows:

[0025] ;

[0026] Where R represents the coding rate dynamically selected based on the real-time channel quality Q, R max Q0 represents the maximum coding rate supported by the system, Q0 represents the channel quality threshold, and k represents the sensitivity coefficient for the steepness of the adjustment function curve.

[0027] Furthermore, in step S6, the graded error handling mechanism includes multi-level thresholds, which are not fixed values ​​but are determined in real time through a dynamic threshold calculation model.

[0028] The construction and calculation process of the dynamic threshold calculation model is as follows: The system continuously monitors and records communication error events occurring within a time window, and collects the time interval sequence of each error event. and the corresponding error severity scoring sequence ;

[0029] First, calculate the statistical coefficient of variation (SCR) of the error time interval sequence within this time window, i.e., the standard deviation of the sequence. with the mean The ratio is used to quantify the discreteness and instability of error occurrence;

[0030] At the same time, calculate the severity score S for all error events within this window. i The square mean is used to amplify the impact of high-frequency or high-severity errors;

[0031] Finally, the coefficient of variation, which represents temporal instability, is multiplied by the mean square, which represents error severity, to obtain a dynamic threshold that comprehensively reflects the current risk level of the communication link. Its calculation formula is defined as:

[0032] ;

[0033] Where N represents the total number of errors within the time window;

[0034] The system uses this dynamic calculation result as the basis for adjusting the trigger sensitivity of the hierarchical error handling mechanism.

[0035] Furthermore, the processing phase of the graded error handling mechanism includes activating an enhanced protection mode for locally stored sensor data;

[0036] The enhanced protection mode includes encrypting the data using a dynamic key generation method based on a chaotic system;

[0037] The generation of the dynamic key fully utilizes the hardware features of the smart helmet. The process involves: real-time acquisition of a series of readings from the device's built-in accelerometer and light sensor within a short period of time, including triaxial acceleration readings a from M sampling points. x ,a y ,a z and ambient light intensity reading (L);

[0038] First, these raw sensor readings are preprocessed by calculating a specific combination of the absolute value of the triaxial acceleration difference and the light intensity reading at each sampling point. The combined results for all sampling points are then summed. To normalize this summation to the interval [0,1) to accommodate the input requirements of subsequent chaotic mapping, a modulo-1 operation is performed on the summation to generate an initial key seed K. seed ;

[0039] The formula for calculating the initial key seed is:

[0040] ;

[0041] in, It is a small constant introduced to avoid division by zero errors;

[0042] Then, this initial seed is input into a Logistic chaotic mapping system, and through multiple nonlinear iterative operations, the initial seed is thoroughly confused and diffused;

[0043] The iterative process of the chaotic mapping is described by the following equation:

[0044] ;

[0045] Among them, K n This represents the chaotic value in the nth iteration. These are parameters that control the behavior of chaotic systems and must ensure that the system is in a chaotic state.

[0046] This process ultimately generates a dynamic encryption key that is highly sensitive to the device’s current physical state and initial conditions.

[0047] Furthermore, the predictive maintenance analysis described in step S7 includes the following steps:

[0048] Collect long-term communication data detection logs, including dynamic fusion check code verification results, channel state parameter sequences, and error handling records;

[0049] Extract multidimensional feature vectors from the logs, including but not limited to verification failure frequency, channel quality fluctuation periodicity, and error type distribution.

[0050] The multidimensional feature vector is input into a pre-trained long short-term memory network model, and the output of the model is a predicted probability value of communication failure occurring within a specific time window in the future.

[0051] If the probability prediction value exceeds the preset alarm threshold, the system will proactively send preventative maintenance suggestions to the monitoring center before the actual fault occurs.

[0052] Furthermore, the training process of the Long Short-Term Memory Network model adopts a spatiotemporal feature joint optimization method. Specifically, in the time dimension, the model learns the periodicity and trend of error events; in the spatial dimension, the model aggregates the communication status of multiple smart safety helmets in the same working area through graph neural networks to capture mutual interference or group failure patterns between devices.

[0053] Furthermore, the visualization and retrospective analysis of communication quality data described in step S8 involves the following steps: the receiving end records the verification results of all data frames, channel state parameters, and triggered error handling actions, along with timestamps, into a distributed database, and provides multi-dimensional queries and visualizations through a web interface. The display formats include, but are not limited to, channel quality heatmaps, bit error rate time series curves, and error type statistical pie charts.

[0054] The beneficial effects of this invention are reflected in:

[0055] By combining payload data, unique device identifiers, and real-time channel state parameters with adaptive weights and the SHA-256 hash algorithm, a dynamic fusion checksum is generated. This enhances the checksum's targeting of high bit error rate dimensions while ensuring the checksum's unpredictability and collision resistance, effectively preventing data tampering or transmission errors. A comprehensive channel quality index is constructed based on Received Signal Strength Indicator (RSSI) and Signal-to-Noise Ratio (SNR). Adaptive coding is achieved through an S-curve rate adjustment function, prioritizing transmission reliability when channel quality is poor and improving transmission efficiency when quality is good, adapting to complex and ever-changing communication environments. The threshold for the graded error handling mechanism is calculated in real-time using a dynamic model, comprehensively considering the occurrence of errors. The system can adjust the processing sensitivity according to the actual risk level of the communication link, avoiding the rigidity of fixed thresholds and accurately responding to different levels of transmission errors, based on the degree of dispersion and severity of the errors. In the advanced error handling stage, a chaotic system dynamic key encryption is used. The key is generated based on real-time data from the device's accelerometer and light sensor, which is extremely sensitive to the device's physical state and initial conditions, greatly improving the encryption strength of locally stored sensor data and preventing data leakage. Through the combination of Long Short-Term Memory (LSTM) network and spatiotemporal feature optimization, predictive maintenance of communication failures is achieved, sending alarms and maintenance suggestions in advance. At the same time, it supports multi-dimensional visualization and retrospective analysis of communication quality data, which facilitates rapid problem location and optimization of communication strategies. Attached Figure Description

[0056] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0057] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0058] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. These embodiments are merely illustrative of the technical solution of the present invention and are therefore intended to limit the scope of protection of the present invention.

[0059] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application should have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0060] like Figure 1 As shown, a communication data detection method based on a smart safety helmet includes the following steps:

[0061] Step S1: Collect multimodal sensor data from the smart safety helmet. The multimodal sensor data includes at least audio data, posture data, and environmental data.

[0062] Step S2: Preprocess the multimodal sensor data, including data alignment, noise reduction and normalization, to form a standardized data stream with a unified timestamp;

[0063] Step S3: Encapsulate the preprocessed standardized data stream into data frames. Each data frame includes a frame header, payload data, and a frame trailer. The frame header includes a sequence number, frame type, and data priority identifier.

[0064] Step S4: Calculate a dynamic fusion check code for each data frame, wherein the dynamic fusion check code is obtained by calculating the payload data, the device unique identifier, and the real-time channel state parameters through a multivariate hash function;

[0065] Step S5: Encode and transmit the data frame using an adaptive coding strategy based on channel state;

[0066] Step S6: At the receiving end, the data frame is received, and the legality of the dynamic fusion check code is verified. If the verification fails, the hierarchical error handling mechanism is adaptively adjusted and executed.

[0067] Step S7: Perform predictive maintenance analysis based on historical logs detected by communication data;

[0068] Step S8: Visualize and retrospectively analyze the communication quality data.

[0069] In step S4, the calculation process of the dynamic fusion check code includes:

[0070] First, the payload data D of the current data frame to be transmitted i The smart helmet's unique hardware identifier (UUID) and channel status parameters (CSI) obtained in real time through the communication module. i As three basic input elements;

[0071] Next, adaptive weight coefficients α, β, and γ are assigned to these three input elements respectively;

[0072] Then, the weighted three element data are subjected to a bitwise XOR operation, with the bitwise XOR operator being ⊕, to achieve the initial fusion of the data and generate a composite feature value;

[0073] Finally, to ensure the unpredictability and collision resistance of this composite feature value, it is input into a preset hash function H for one-way hash calculation. This hash function uses the SHA-256 algorithm, and the mathematical expression of the calculation process is as follows:

[0074] ;

[0075] in, This represents the final dynamically fused checksum of the i-th frame, which is closely related to the current communication context.

[0076] By using the payload data of the data frame to be transmitted, the unique hardware identifier (UUID) of the smart safety helmet, and the real-time channel state parameters (CSIi) as verification inputs, and combining adaptive weight allocation with the SHA-256 hash algorithm to generate a dynamic fusion check code, multi-dimensional verification of "data itself + device identity + channel environment" is achieved, avoiding the vulnerabilities of single-element verification (such as only verifying the payload). Adaptive weights can also enhance the verification contribution of high bit error rate dimensions. At the same time, the SHA-256 algorithm ensures the unpredictability and collision resistance of the check code, which can effectively prevent data tampering, forgery, or misjudgment caused by channel fluctuations, and significantly improve the accuracy and security of communication data verification.

[0077] Setting the scenario and basic parameters: Taking a construction scenario as an example, a smart safety helmet needs to transmit environmental data (temperature 29℃, humidity 62%), and the corresponding load data D is... i The encoding is binary 10110010 (decimal 178); the device's unique hardware identifier (UUID) is hexadecimal 6A4B8C2D (converted to decimal 1783450669); the channel state parameter (CSI) is obtained in real time through the communication module. i (Signal-to-noise ratio 22dB), encoded as binary 11010100 (decimal 2^12).

[0078] Determine adaptive weights: Based on the system's statistics of historical communication data over the past hour, the load data D i The bit error rate is 4% (highest bit error rate), the bit error rate of UUID is 0.08% (extremely low bit error rate), and CSI... i The bit error rate is 1.8% (medium bit error rate), therefore, following the strategy of "higher weight for higher bit error rate dimensions," an adaptive weight α = 0.65 (corresponding to D) is assigned. i ), β=0.1 (corresponding to UUID), γ=0.25 (corresponding to CSI) i ).

[0079] Calculate the weighted data: Weight the three elements according to their respective weights, and the results are as follows:

[0080] Weighted load data: α⋅D i =0.65×178=115.7 (binary 1110011.101);

[0081] Weighted UUID: β⋅UUID = 0.1 × 1783450669 = 178345066.9 (binary 10101010011000101101100010101.110011...).

[0082] Weighted channel state parameter: γ⋅CSI i =0.25×2^12=53 (binary 110101).

[0083] Perform a bitwise XOR operation: Perform a bitwise XOR (⊕) operation on the weighted three-element data to achieve initial fusion and obtain the composite feature value:

[0084] 1110011.101⊕10101010011000101101100010101.110011...⊕110101=10101010011000101101100100010.011001... (To simplify subsequent calculations, the integer part is taken as binary: 10101010011000101101100100010).

[0085] Generate a dynamic fusion checksum: Input the above composite feature values ​​into the preset SHA-256 hash function (H), and generate a dynamic fusion checksum according to the formula. The dynamic fusion check code is calculated, and the result is hexadecimal "7E2A9D4F1C8B3G5H6J7K8L9M0N1P2Q3R4" (the actual SHA-256 calculation result is a 64-bit hexadecimal value, which is simplified for example).

[0086] Verify the effectiveness of tamper detection: If, during transmission, the payload data D... i Because the electromagnetic interference at the construction site was altered to binary 10110000 (decimal 176), the calculation was repeated according to the steps above: α⋅D i With β⋅UUID, γ⋅CSI i The XOR operation yields a new composite feature value, which is then used in SHA-256 calculation to obtain the new checksum "5C8B2E7D1F3A4H6J5K9L8M0N2P1Q3R4". This new checksum is identical to the original... Completely different, the receiving end can accurately identify data tampering; at the same time, due to the integration of UUID, if a counterfeit device uses another UUID to send data, even if the payload matches the CSI, the verification code will fail, further ensuring the legitimacy of the data source.

[0087] The adaptive weight coefficients α, β and γ are not fixed values, but are dynamically adjusted after performing sliding window statistical analysis on historical communication success rate and data integrity indicators. The adjustment strategy is as follows: the higher the historical error rate of the data dimension, the larger the corresponding weight coefficient is assigned in the next period, so as to enhance the contribution of the verification dimension in the overall verification.

[0088] By using a sliding window to statistically analyze historical communication success rates and data integrity indicators, the adaptive weights α, β, and γ corresponding to payload data, UUID, and channel state parameters are dynamically adjusted. This gives greater weight to data dimensions with high bit error rates, avoiding the rigidity of low bit error rate dimensions "masking" problems in high bit error rate dimensions under fixed weights. This ensures that the verification focus always matches the actual data quality fluctuations, significantly improving the sensitivity of the dynamic fusion check code to high-risk data dimensions, further enhancing the targeting and accuracy of the verification, and ensuring that the verification effect remains reliable under different communication environments.

[0089] Continuing with the smart safety helmet used in construction scenarios (UUID remains hexadecimal 6A4B8C2D, decimal 1783450669), the sliding window duration is set to 10 minutes, and the load data D within this window is statistically analyzed. i (Ambient temperature and humidity data), device UUID, channel status parameter CSI i The total number of transmissions and the number of bit errors (signal-to-noise ratio) are considered. The total data transmission frequency within the window is 500 times (1 transmission every 1.2 seconds).

[0090] Historical communication data statistics and bit error rate calculation: Through sliding window analysis, the transmission data of three elements are obtained: ①D i Total transmissions: 500, with 28 errors due to construction site dust interference; error rate: ;

[0091] ②UUID: Total transmissions: 500, with only 1 error due to occasional hardware fluctuations, resulting in an error rate of [missing information]. ;

[0092] CSI i Total transmissions: 500, with 15 errors due to electromagnetic interference from the tower crane. Error rate: Bit error rate sorted as follows .

[0093] The system uses an adjustment strategy of "the higher the bit error rate, the greater the weight," allocating weights based on sliding window statistics (the total weights are 1): because R d The highest allocation is α=0.7; R CSI Next, allocate γ=0.25; R UUID The lowest value is assigned β=0.05 (for fixed-weight schemes, α=β=γ≈0.33 is usually assumed).

[0094] Comparison of verification performance in error scenarios: D in simulated transmission i One bit error occurred (original binary data 10110010 → altered to 10110000, decimal 178 → 176), CSI i Scenarios where there are no errors with UUID:

[0095] ① Calculate the verification contribution using dynamic weights: α⋅D i The contribution of (error) is 0.7, γ⋅CSI i (Correct) is 0.25, β⋅UUID (correct) is 0.05, the error dimension contributes 70% of the total, and the checksum can quickly capture D. i mistake;

[0096] ② Calculate the error dimension using fixed weights (α=β=γ≈0.33). i With a contribution of only 33%, the "high weight" of the correct dimension can easily offset erroneous signals, leading to verification delays or misjudgments. A comparison of the two weighting methods shows that dynamically adjusted weights allow the verification focus to be placed on high-error-rate components. i This avoids the blind spots of fixed-weight verification and improves the accuracy of error identification.

[0097] In step S5, the adaptive coding strategy based on channel state specifically includes:

[0098] First, the two key physical layer parameters, Received Signal Strength Indicator (RSSI) and Signal-to-Noise Ratio (SNR), are obtained in real time through the underlying driver of the communication module, and then linearly combined into a normalized comprehensive channel quality index, Q.

[0099] Then, a rate adjustment function is set. The core feature of this function is that when the channel quality is below a certain threshold, the coding rate can be gradually reduced to prioritize reliability, while when the channel quality is good, the coding rate can be quickly increased to pursue transmission efficiency.

[0100] This function uses an S-shaped curve to simulate this adaptive adjustment behavior, and its mathematical expression is used to accurately calculate the optimal coding rate to be selected under the current channel conditions. The expression is as follows:

[0101] ;

[0102] Where R represents the coding rate dynamically selected based on the real-time channel quality Q, R max Q0 represents the maximum coding rate supported by the system, Q0 represents the channel quality threshold, and k represents the sensitivity coefficient for the steepness of the adjustment function curve.

[0103] The underlying driver of the communication module acquires RSSI (Received Signal Strength Indicator) and SNR (Signal-to-Noise Ratio) in real time, and linearly combines the two into a normalized comprehensive channel quality index Q. Then, based on the S-curve rate adjustment function, the optimal coding rate is dynamically calculated. This achieves a smooth rate reduction to prioritize reliability when the channel quality is poor, and a rapid rate increase to pursue transmission efficiency when the quality is good. This avoids the problem of packet loss or bandwidth waste caused by a fixed coding rate in complex environments (such as electromagnetic / obstruction interference in building construction scenarios). It accurately balances communication reliability and efficiency and adapts to dynamically changing channel conditions.

[0104] For example, in a smart safety helmet used in construction, the maximum encoding rate R supported by the system is set. max =1Mbps, channel quality threshold Q0=0.5 (after normalization, 0 is the worst and 1 is the best), adjustment function sensitivity coefficient k=10 (controlling the steepness of the S-curve); at the same time, the normalization calculation method of the comprehensive channel quality index Q is defined:

[0105] ;

[0106] RSSI min =-120dBm (weakest signal), RSSI max =-30dBm (strongest signal), SNR min =0dB (no valid signal), SNR max =30dB (optimal signal-to-noise ratio).

[0107] Scenario 1: Poor channel quality (tower crane obstruction + electromagnetic interference): The smart safety helmet is placed near the tower crane operating area, and RSSI=-100dBm and SNR=5dB are obtained in real time; first calculate Q:

[0108] ;

[0109] Substituting back into the S-shaped rate formula: (Approximately 43kbps) At this speed, the data rate is low, but it can prioritize ensuring that data is not lost and avoid transmission failure due to poor channel conditions.

[0110] Scenario 2: Medium channel quality (no obstruction at the edge of the construction area): The smart safety helmet is moved to the edge of the construction area, with no tower crane obstruction, and RSSI=-70dBm and SNR=15dB are obtained in real time; calculate Q:

[0111] ;

[0112] Substitute into the rate formula: At this point, the speed is moderate, balancing reliability and transmission efficiency, and meeting the needs for real-time transmission of temperature and humidity data.

[0113] Scenario 3: Excellent Channel Quality (Near Monitoring Center): The smart safety helmet is located near the project's monitoring center, with no interference or obstruction, and real-time RSSI=-40dBm and SNR=25dB are obtained; Q is calculated.

[0114] ;

[0115] Substitute into the rate formula: (close to R) max At this point, the speed is maximized, data is transmitted quickly, bandwidth resources are not wasted, and reliability is still guaranteed due to the excellent channel.

[0116] In step S6, the graded error handling mechanism includes multi-level thresholds, which are not fixed values ​​but are determined in real time through a dynamic threshold calculation model.

[0117] The construction and calculation process of the dynamic threshold calculation model is as follows: The system continuously monitors and records communication error events occurring within a time window, and collects the time interval sequence of each error event. and the corresponding error severity scoring sequence ;

[0118] First, calculate the statistical coefficient of variation (SCR) of the error time interval sequence within this time window, i.e., the standard deviation of the sequence. with the mean The ratio is used to quantify the discreteness and instability of error occurrence;

[0119] At the same time, calculate the severity score S for all error events within this window. i The square mean is used to amplify the impact of high-frequency or high-severity errors;

[0120] Finally, the coefficient of variation, which represents temporal instability, is multiplied by the mean square, which represents error severity, to obtain a dynamic threshold that comprehensively reflects the current risk level of the communication link. Its calculation formula is defined as:

[0121] ;

[0122] Where N represents the total number of errors within the time window;

[0123] The system uses this dynamic calculation result as the basis for adjusting the trigger sensitivity of the graded error handling mechanism;

[0124] The threshold of the graded error handling mechanism is calculated in real time through a dynamic model. This model integrates the discrete instability of error occurrence within a time window (the coefficient of variation of the error time interval) and the severity of the error (the squared mean of the error severity). This avoids the rigidity of fixed thresholds that cannot adapt to real-time risk changes in communication links. It can accurately reflect the current risk level of the link, making the trigger sensitivity of error handling match the actual risk. This prevents over-processing and waste of resources when the risk is low, and also avoids data loss or communication interruption due to untimely processing when the risk is high, thus improving the accuracy and adaptability of error handling.

[0125] For example, in a smart safety helmet used in a construction site (which transmits ambient temperature and humidity data), the time window for calculating the dynamic threshold is set to 10 minutes (600 seconds). Within this window, the system records 5 communication error events (referred to as events 1-5) and collects the error time interval sequence (starting from the first error, the time difference between adjacent errors). The corresponding error severity rating sequence (1-5 points, 1 for minor packet loss, 5 for complete data corruption) The total number of errors is N=5.

[0126] Calculate the statistical coefficient of variation for the error time interval:

[0127] The first step is to calculate the mean of the error time intervals. :

[0128] ;

[0129] The second step is to calculate the standard deviation of the error time interval. First, calculate the sum of squares of the deviations of each interval from the mean:

[0130] ;

[0131] Next, calculate the standard deviation: ;

[0132] The third step is to calculate the coefficient of variation. This value reflects that errors occur in a dispersed (non-uniform) manner, indicating poor link stability.

[0133] Calculate the squared mean of error severity: according to the formula Substituting the data, we get:

[0134] This value amplifies the impact of high-severity errors (such as S=5 in event 2) due to the squaring operation, reflecting the overall high harm of the error.

[0135] Calculate dynamic threshold According to the formula Substituting the data, we get: .

[0136] Comparison of processing effects between dynamic thresholds and fixed thresholds: If a fixed threshold (constantly set) is used... ),at this time Fixed thresholds may misjudge the current link risk as low and only trigger low-level error handling (such as simple retransmission). However, in reality, due to the dispersion of errors and the high proportion of high-severity errors, low-level handling cannot solve the problem and may lead to continuous loss of subsequent data.

[0137] Dynamic threshold It accurately reflects high risks and triggers advanced error handling (such as switching to backup channels and local data encryption storage), effectively avoiding risks. It can be seen that dynamic thresholds can accurately match the actual risks of the link and avoid the processing deviations of fixed thresholds.

[0138] The processing phase of the graded error handling mechanism includes activating an enhanced protection mode for locally stored sensor data;

[0139] The enhanced protection mode includes encrypting the data using a dynamic key generation method based on a chaotic system;

[0140] The generation of the dynamic key fully utilizes the hardware features of the smart helmet. The process involves: real-time acquisition of a series of readings from the device's built-in accelerometer and light sensor within a short period of time, including triaxial acceleration readings a from M sampling points. x ,a y ,a z and ambient light intensity reading (L);

[0141] First, these raw sensor readings are preprocessed by calculating a specific combination of the absolute value of the triaxial acceleration difference and the light intensity reading at each sampling point. The combined results for all sampling points are then summed. To normalize this summation to the interval [0,1) to accommodate the input requirements of subsequent chaotic mapping, a modulo-1 operation is performed on the summation to generate an initial key seed K. seed ;

[0142] The formula for calculating the initial key seed is:

[0143] ;

[0144] in, It is a small constant introduced to avoid division by zero errors;

[0145] Then, this initial seed is input into a Logistic chaotic mapping system, and through multiple nonlinear iterative operations, the initial seed is thoroughly confused and diffused;

[0146] The iterative process of the chaotic mapping is described by the following equation:

[0147] ;

[0148] Among them, K n This represents the chaotic value in the nth iteration. These are parameters that control the behavior of chaotic systems and must ensure that the system is in a chaotic state.

[0149] This process ultimately generates a dynamic encryption key that is extremely sensitive to the device’s current physical state and initial conditions.

[0150] The advanced stage of graded error handling encrypts local sensor data using dynamic keys based on a chaotic system. The key generation fully utilizes real-time hardware data from the smart helmet's accelerometer and light sensor, and after iterative chaotic mapping, it becomes extremely sensitive to the device's physical state (such as the wearer's actions) and initial conditions. Compared to fixed keys or general encryption methods, it is much more difficult to crack. It can provide strong dedicated encryption protection for critical data (such as environmental and attitude data) stored locally when the communication link is at high risk (such as frequent verification failures), preventing data from being illegally stolen or tampered with, and significantly enhancing local data security in advanced error scenarios.

[0151] Unified scenario and parameter settings: Continuing with the smart safety helmet from the construction scenario (advanced error handling has been triggered previously, requiring the preservation of nearly 1 minute of locally stored environmental / attitude data), the following parameters are set: Number of sensor sampling points M=5 (sampling once every 0.2 seconds, for a total of 1 second), avoiding small constants that could lead to division-by-zero errors. Control parameters of Logistic chaotic mapping (Ensure the system is in a chaotic state);

[0152] The sensor data from the 5 sampling points are as follows:

[0153] Sampling point 1: ;

[0154] Sampling point 2: ;

[0155] Sampling point 3: ;

[0156] Sampling point 4: ;

[0157] Sampling point 5: .

[0158] Calculate the initial key seed K seed According to the formula Calculate and sum point by point:

[0159] Sampling point 1: ;

[0160] Sampling point 2: ;

[0161] Sampling point 3: ;

[0162] Sampling point 4: ;

[0163] Sampling point 5: ;

[0164] Sum: 0.00871 + 0.01082 + 0.00438 + 0.00873 + 0.00645 ≈ 0.03909;

[0165] Modulo-1 operation: .

[0166] Generating a dynamic key using Logistic chaotic mapping: Substituting into the chaotic mapping formula The final 16-bit binary key is generated through 5 iterations (the number of iterations can be adjusted as needed; 5 iterations are used here):

[0167] First iteration: ;

[0168] Second iteration: ;

[0169] 3rd iteration: ;

[0170] 4th iteration: ;

[0171] 5th iteration: ;

[0172] Will Converted to 16-bit binary (keeping the last 16 decimal places): 0.0011111010111000, this is the final dynamic encryption key.

[0173] Safety comparison verification: If worker movement causes a change in the posture of the safety helmet (such as at sampling point 3) From 0.9m / s 2 It becomes 1.6 m / s 2 ), recalculate K seed :

[0174] Sampling point 3 The calculation results for the remaining sampling points remain unchanged, and the sum is approximately 0.03909 - 0.00438 + 0.01973 = 0.05444, K. seed =0.05444mod1=0.05444;

[0175] After chaotic iteration, the 5th iteration Converted to 16-bit binary, this is 0.0010111100010110, which is significantly different from the original key. However, if a fixed key (such as 0.0011001100110011) is used, it remains unchanged regardless of the device's physical state, making it easily cracked. Therefore, dynamic keys, because they are bound to the device's real-time physical state, offer far greater security than fixed keys and can effectively protect local data in advanced error scenarios.

[0176] The predictive maintenance analysis described in step S7 includes the following steps:

[0177] Collect long-term communication data detection logs, including dynamic fusion check code verification results, channel state parameter sequences, and error handling records;

[0178] Extract multidimensional feature vectors from the logs, including but not limited to verification failure frequency, channel quality fluctuation periodicity, and error type distribution.

[0179] The multidimensional feature vector is input into a pre-trained long short-term memory network model, and the output of the model is a predicted probability value of communication failure occurring within a specific time window in the future.

[0180] If the probability prediction value exceeds the preset alarm threshold, the system will proactively send preventative maintenance suggestions to the monitoring center before the actual fault occurs.

[0181] By collecting long-term communication data detection logs, extracting multi-dimensional feature vectors such as verification failure frequency, channel quality fluctuation periodicity, and error type distribution, and inputting them into a pre-trained Long Short-Term Memory (LSTM) network model, the probability of communication failure within a specific time window in the future can be accurately predicted. This enables a shift from "passively handling failures" to "actively preventing failures," avoiding data loss or communication interruptions caused by sudden failures. At the same time, preventive maintenance suggestions can be sent in advance, reducing downtime caused by fault diagnosis and repair, and significantly improving the operation and maintenance efficiency and reliability of the smart safety helmet communication system.

[0182] For example, in a construction site, a smart safety helmet (transmitting data on ambient temperature and humidity, and wearer posture) collects communication data detection logs from the past 30 days. Daily data records include: the number of dynamic fusion checksum verification failures, the hourly channel quality index Q (calculated based on RSSI and SNR) sequence, and error type records (checksum error, encoding error, transmission timeout). The prediction target is set as "the probability of communication failure during the period from 8:00 to 12:00 in the next 24 hours (day 31)". The LSTM model has been pre-trained using 180 days of historical data, and a preset failure probability alarm threshold P is set. th =0.7.

[0183] Multidimensional feature vector extraction: Extract three core features from the logs of the past 30 days to construct the feature vector input to the LSTM. :

[0184] Feature 1 (Verification Failure Frequency X1): Calculates the average daily number of verification failures over the past 30 days, using the following formula: F d This represents the number of verification failures on day d; the log shows a total of 285 failures over 30 days, therefore... (times / day).

[0185] Feature 2 (Channel Quality Fluctuation Periodicity X2): Fourier transform was performed on 24 Q values ​​(1 per hour) for each day of the past 30 days to extract the period corresponding to the fluctuation peak; the results showed that the Q value fluctuation was the largest from 10:00 to 14:00 (peak tower crane operation) each day, with a peak period T=4 hours, so X=4 (hours).

[0186] Feature 3 (Error Type Distribution X3): Calculate the percentage of each type of error over the past 30 days, using the following formula: (Since verification errors accounted for the highest percentage, they were chosen as representative); the log showed a total of 420 errors, with 252 verification errors. .

[0187] The final feature vector is .

[0188] LSTM Model Prediction and Threshold Judgment: The feature vector X is input into a pre-trained LSTM model, and the model outputs a predicted value P for the communication failure probability during the next 24 hours from 8:00 to 12:00. pred =0.85; compared to the alarm threshold P th =0.7, because The system determined that there was a high risk of failure during that period.

[0189] Early warning and maintenance effectiveness verification: The system immediately sent preventative maintenance recommendations to the monitoring center, including "checking for communication signal interference sources in the tower crane operation area during the 8:00-12:00 period and replacing the communication module antenna of the smart safety helmet in advance." After the maintenance personnel followed the recommendations, the actual number of failures during the 8:00-12:00 period on the 31st day was only 1 (compared to the original prediction of about 6 failures without maintenance), and the communication normality rate increased from the predicted 65% to 98%. If this predictive mechanism had not been adopted, a 2-hour work stoppage would have been required to investigate the interference source after a sudden failure, which would have caused delays in the construction progress. Predictive maintenance avoided this loss, demonstrating its value in improving operation and maintenance efficiency.

[0190] The training process of the Long Short-Term Memory Network model adopts a spatiotemporal feature joint optimization method. Specifically, in the time dimension, the model learns the periodicity and trend of error events; in the spatial dimension, the model aggregates the communication status of multiple smart safety helmets in the same working area through graph neural networks to capture mutual interference or group failure patterns between devices.

[0191] The training of the Long Short-Term Memory (LSTM) network model employs joint optimization of spatiotemporal features. In the temporal dimension, it can accurately learn the periodicity (such as the interference cycle during peak tower crane operations in construction scenarios) and trend (such as the upward trend of error rate due to equipment aging), avoiding spatial correlation problems that are easily overlooked when relying solely on the time dimension. In the spatial dimension, it aggregates the communication status of multiple smart safety helmets in the same work area through graph neural networks, which can capture group correlations that cannot be reflected by a single device (such as multiple devices in the same area being subject to common electromagnetic interference or base station signal coverage blind spots, resulting in group failures). This solves the limitations of training with data from a single device, ultimately making the model's prediction of communication failures more comprehensive and accurate. In particular, it significantly improves the recognition rate of periodic and group failures, reduces prediction omissions or misjudgments, and further enhances the reliability of predictive maintenance.

[0192] RU construction scene, focusing on construction area A of the project department (approximately 500m²). 2 (Including 3 tower cranes), 5 smart safety helmets (numbered S1-S5, all transmitting environmental temperature, humidity and attitude data) are deployed in this area; the model training data is the communication logs of the past 180 days, with the time dimension divided into 1-hour units and the spatial dimension within the construction area A. S1-S5 are regarded as 5 nodes (devices) of a graph neural network, and the correlation weight between nodes is based on the distance between devices (the smaller the distance, the greater the weight, simulating the spatial correlation of signal interference).

[0193] Time-dimensional feature learning process: For 180 days of logs from a single device (e.g., S1), extract the error count sequence for each time period of the day (a total of 24 time periods). (d is the number of days, t is the time period, ranging from 1 to 24), periodicity is calculated using the autocorrelation function:

[0194] The formula for the autocorrelation coefficient is: ,in Let τ be the average number of errors in S1 and τ be the time lag. When τ = 4 (maximum correlation coefficient), it is found that there is a 4-hour cycle in the error events of S1 (corresponding to two peak tower crane operation times of 10:00-14:00 and 16:00-20:00 each day). The model successfully learned this periodicity in the time dimension. At the same time, by fitting the error rate trend through linear regression, it is found that the error rate of S1 has increased by an average of 0.8% per month in the past 30 days, capturing the trend of equipment antenna aging.

[0195] Spatial Dimension Graph Neural Network Aggregation Process: Constructing the adjacency matrix A (a 5×5 matrix) of the graph neural network. (The association weights between devices Si and Sj), the weights are calculated based on the device spacing:

[0196] ( (where S is the straight-line distance between Si and Sj, in meters).

[0197] Given that the spacing between S1 and S2 is 15m, S1 and S3 is 30m, S1 and S4 is 25m, S1 and S5 is 20m, S2 and S3 are 20m, S2 and S4 are 35m, S2 and S5 are 25m, S3 and S4 are 18m, S3 and S5 are 22m, and S4 and S5 are 12m, we can obtain some elements of the adjacency matrix:

[0198] ;

[0199] The real-time channel quality Q-values ​​of the five devices (S1=0.3, S2=0.28, S3=0.45, S4=0.42, S5=0.38) were used as node features, and the region features were aggregated using mean pooling in a graph neural network.

[0200] It was found that the overall Q value of region A was low (below the threshold of 0.5), indicating a group-wide weak signal problem. The Q values ​​of individual devices S3 and S4 (0.45 and 0.42) were easily misjudged as "medium quality" when viewed alone, and the regional risk was only exposed after spatial aggregation.

[0201] Comparison of spatiotemporal joint training and prediction performance: Comparison of two model training schemes:

[0202] ① Training only in the time dimension (using only time features from a single device);

[0203] ② Spatiotemporal joint training (integrating temporal features and spatial aggregation features);

[0204] The fault prediction results for construction area A in the next 24 hours show:

[0205] Option ① has an 82% accuracy rate in predicting periodic faults (10:00-14:00), but it misses the prediction of group faults (synchronization error of all equipment in area A from 11:00-13:00).

[0206] Option 2 improves the accuracy of periodic fault prediction to 95% and accurately identifies group faults (prediction accuracy of 92%). When a fault actually occurs, Option 2 provides a 25-minute lead time compared to Option 1, allowing more time for maintenance and fully demonstrating the role of spatiotemporal joint optimization in enhancing the model's predictive capabilities.

[0207] The steps for visualizing and retrospectively analyzing communication quality data in step S8 are as follows: The receiving end records the verification results of all data frames, channel state parameters, and triggered error handling actions, along with timestamps, into a distributed database, and provides multi-dimensional queries and visualizations through a web interface. The display formats include, but are not limited to, channel quality heatmaps, bit error rate time series curves, and error type statistical pie charts.

[0208] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A communication data detection method based on a smart safety helmet, characterized in that: Includes the following steps: Step S1: Collect multimodal sensor data from the smart safety helmet; Step S2: Preprocess the multimodal sensor data to form a standardized data stream with a unified timestamp; Step S3: Encapsulate the preprocessed standardized data stream into a data frame; Step S4: Calculate the dynamic fusion check code for each data frame. The dynamic fusion check code is obtained by calculating the payload data, the device unique identifier, and the real-time channel state parameters through a multivariate hash function. Step S5: Encode and transmit the data frame using an adaptive coding strategy based on channel state; Step S6: At the receiving end, the data frame is received, and the legality of the dynamic fusion check code is verified. If the verification fails, the hierarchical error handling mechanism is adaptively adjusted and executed. Step S7: Perform predictive maintenance analysis based on historical logs detected by communication data; Step S8: Visualize and retrospectively analyze the communication quality data.

2. The communication data detection method based on a smart safety helmet according to claim 1, characterized in that: In step S4, the calculation process of the dynamic fusion check code includes: First, the payload data of the current data frame to be transmitted, the unique hardware identifier UUID of the smart safety helmet, and the channel state parameters obtained in real time through the communication module are taken as three input elements. Next, adaptive weight coefficients are assigned to each of the three input elements; Then, the weighted three element data are XORed bitwise to achieve the initial data fusion and generate a composite feature value. Finally, to ensure the unpredictability and collision resistance of the composite feature value, it is input into a preset hash function for one-way hash calculation, which uses the SHA-256 algorithm.

3. The communication data detection method based on a smart safety helmet according to claim 2, characterized in that: The adaptive weight coefficient is not a fixed value, but is dynamically adjusted after performing sliding window statistical analysis on historical communication success rate and data integrity indicators. The adjustment strategy is as follows: the higher the historical error rate of the data dimension, the larger the corresponding weight coefficient is assigned in the next period to enhance the contribution of the verification dimension in the overall verification.

4. The communication data detection method based on a smart safety helmet according to claim 3, characterized in that: In step S5, the adaptive coding strategy based on channel state specifically includes: First, the two key physical layer parameters, Received Signal Strength Indicator (RSSI) and Signal-to-Noise Ratio (SNR), are obtained in real time through the underlying driver of the communication module, and then linearly combined into a normalized comprehensive channel quality index, Q. Then, a rate adjustment function is set. The rate adjustment function can reduce the coding rate when the channel quality is at a certain preset threshold, and increase the coding rate when the channel quality is good. The rate adjustment function uses an S-shaped curve to simulate adaptive adjustment behavior.

5. The communication data detection method based on a smart safety helmet according to claim 4, characterized in that: In step S6, the graded error handling mechanism includes multi-level thresholds, which are not fixed values ​​but are determined in real time through a dynamic threshold calculation model. The construction and calculation process of the dynamic threshold calculation model is as follows: The system continuously monitors and records communication error events that occur within a time window, and collects the time interval sequence of each error event. and the corresponding error severity scoring sequence ; First, calculate the statistical coefficient of variation of the error time interval sequence within the time window, which is the ratio of the standard deviation of the sequence to the mean, to quantify the dispersion and instability of the error occurrence. At the same time, the squared mean of the severity scores of all error events within the window is calculated to amplify the impact of high-frequency or high-severity errors; Finally, the coefficient of variation, which represents temporal instability, is multiplied by the mean square, which represents the severity of the error, to obtain a dynamic threshold that comprehensively reflects the current risk level of the communication link. The system uses this dynamically calculated dynamic threshold as the basis for adjusting the trigger sensitivity of the graded error handling mechanism.

6. The communication data detection method based on a smart safety helmet according to claim 5, characterized in that: The processing phase of the graded error handling mechanism includes activating an enhanced protection mode for locally stored sensor data; The enhanced protection mode includes encrypting the data using a dynamic key generation method based on a chaotic system; The dynamic key generation process of the smart safety helmet's hardware features involves: real-time acquisition of readings from the device's built-in accelerometer and light sensor over a short period, specifically the absolute value of the acceleration difference and the light intensity reading, including triaxial acceleration readings a from M sampling points. x ,a y ,a z and ambient light intensity reading (L); First, the readings collected in the above process are preprocessed, that is, a specific combination value is obtained by taking the absolute value of the acceleration difference and the light intensity reading at each sampling point; Then, the specific combination values ​​of all M sampling points are summed. In order to normalize the summation result to the interval [0,1) to meet the input requirements of subsequent chaotic mapping, a modulo-1 operation is performed on the summation result to generate the initial key seed. The initial key seed is input into the Logistic chaotic mapping system. Through multiple nonlinear iterative operations, the initial seed is thoroughly confused and diffused, ultimately generating a dynamic encryption key for the current physical state of the device.

7. The communication data detection method based on a smart safety helmet according to claim 1, characterized in that: The predictive maintenance analysis described in step S7 includes the following steps: Collect long-term communication data detection logs, including dynamic fusion check code verification results, channel state parameter sequences, and error handling records; Extract multidimensional feature vectors from the logs, including but not limited to verification failure frequency, channel quality fluctuation periodicity, and error type distribution; The multidimensional feature vector is input into a pre-trained long short-term memory network model, and the output of the model is a predicted probability value of communication failure occurring within a specific time window in the future. If the probability prediction value exceeds the preset alarm threshold, the system will proactively send preventative maintenance suggestions to the monitoring center before the actual fault occurs.

8. The communication data detection method based on a smart safety helmet according to claim 7, characterized in that: The training process of the Long Short-Term Memory Network model adopts a spatiotemporal feature joint optimization method, specifically: in the time dimension, the model learns the periodicity and trend of error events; In the spatial dimension, the model aggregates the communication status of multiple smart safety helmets within the same working area through graph neural networks to capture mutual interference or collective failure modes between devices.

Images