A mine equipment fault early warning system and method based on multi-source data

Abstract

The application discloses a mine equipment fault early warning system and method based on multi-source data, and relates to the technical field of fault prediction. The application synchronously collects three original signals of vibration signals, temperature signals and current load rate signals of mine equipment during work by monitoring points, and deducts zero point slow drift components from the original signals to obtain raw signals; according to the current temperature and the current load rate, the characteristic fingerprint vector of the corresponding vibration signal is found, and the current raw signal is normalized; the normal energy envelope line is calculated by using the energy of the vibration signal of the mine equipment during the historical normal work, the residual energy ratio is calculated, and the fault sign activation is determined by using the residual energy ratio; the sign phase latch is established, the residual energy ratio is used for iterative calculation on the sign phase latch, and whether the fault sign is an effective sign is judged by using the iterative result; when the effective sign is judged, the current residual energy difference is finally judged by using the early warning threshold to trigger the early warning.

Description

Technical Field

[0001] This invention relates to the field of fault prediction technology, specifically to a fault early warning system and method for mining equipment based on multi-source data. Background Technology

[0002] Mining equipment operates under harsh conditions of high dust, high humidity, strong vibration, and variable loads for extended periods. The health of its key components directly affects the continuity and safety of mine production. In recent years, with the development of sensor technology and the Industrial Internet of Things (IIoT), equipment fault early warning systems based on multi-source data such as vibration, temperature, and current have become an important means of intelligent operation and maintenance for mining equipment.

[0003] Existing technologies commonly face the dilemma of high false alarm rates and high false negative rates in practical mine applications. Extensive field statistics show that the false alarm rate of existing systems often exceeds 30%, while the false negative rate for early, minor faults, such as bearing micro-wear, hydraulic internal leakage, and early gear pitting, remains high. In-depth analysis generally attributes this to the complex and variable nature of mine conditions or the insufficient generalization ability of deep learning models. However, this overlooks a more hidden root cause: the signal preprocessing stage lacks the ability to effectively distinguish and separate three easily confused types of interference: slow zero-point drift of sensors, inherent fluctuations in operating conditions, and the subtle trend characteristics of real early faults. Summary of the Invention

[0004] The purpose of this invention is to provide a mining equipment fault early warning system and method based on multi-source data to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for early warning of mining equipment faults based on multi-source data, the method comprising the following steps: S100. Sensors are set up in the mining equipment to form monitoring points. The monitoring points are used to synchronously collect three raw signals: vibration signal, temperature signal and current load rate signal when the mining equipment is working. The signal time axis is divided into three semantic segments: steady state segment, transient segment and transition segment. Furthermore, the specific steps for dividing the signal time axis into three semantic segments—steady-state segment, transient segment, and transition segment—are as follows: S101. Calculate the rate of change R of the current load rate signal of the mining equipment. L =dL(t) / dt, where L(t) represents the current load rate of the mining equipment; the vibration signal represents the vibration acceleration of the equipment; Preset noise thresholds δ1 and impulse thresholds δ2, in the signal time axis when |R L When | ≤ δ1, it is divided into a steady state segment; when δ1 < |R LWhen | < δ2, it is divided into a transition segment; when | R L When |≥δ2, it is divided into transient segments.

[0006] The steady-state segment indicates that the load rate change is completely submerged in sensor noise and can be regarded as zero change; the transition segment indicates that it is neither equal to noise nor equal to impact, covering common scenarios in mines such as gradual change in tunneling hardness and slow accumulation of material on conveyor belts; the transient segment indicates equipment startup, emergency shutdown, and sudden impact of the bucket against the rock wall, which will be given lower weight in subsequent steps. The three original signals provide physical "time anchors" for the separation of different signal components, avoiding confusion between normal transient behaviors such as startup and shutdown and sensor drift or equipment failure, thus reducing the false alarm rate from the source.

[0007] The introduction of a "transition section" covers scenarios such as gradual changes in tunneling hardness and slow material accumulation, filling the gap in traditional methods that only distinguish between "steady state / transient state". This allows the unique gradual working conditions of mines to be preserved and given reasonable weight, avoiding misjudgment of gradual working conditions as faults.

[0008] S200. Set a sliding freeze window, perform median fitting of the steady-state segment of the vibration signal to form a slow drift baseline function, take the vibration signal value of the slow drift baseline function as the zero-point slow drift component, and subtract the zero-point slow drift component from the original signal to obtain the gross signal. Furthermore, the specific steps for subtracting the slow zero-point drift component from the original signal to obtain the gross signal are as follows: S201, The length of the sliding freeze window is T. freeze The time frame can be 72 hours, with a step size of 1 hour. Within each sliding freeze window, the median M of the vibration signals in all steady-state segments is extracted. Over n consecutive sliding freeze windows, the median sequence {M1, M2, M3...M...} is generated. n The anchor point trajectories, M1, M2, M3...M, are considered as sensor drift. n This represents the median of the vibration signal for the 1st, 2nd, 3rd...nth consecutive sliding freeze window; By utilizing the physical characteristic of "extremely slow" drift, the zero-point drift that takes months to separate from the vibration signal that takes seconds or minutes, thus avoiding the damage to low-frequency fault components caused by traditional high-pass filtering.

[0009] S202. Fit the anchor point trajectory using the minimum energy path constraint to obtain the slow drift baseline function D(t); the minimum energy path constraint includes the first derivative boundedness and the second derivative smoothing, the first derivative boundedness means |dD(t) / dt|<ε, ε represents the maximum allowable drift rate, which is determined by the sensor manual; The second derivative smoothing means that the second derivative of D(t) takes a minimum value, which is set by the staff; this ensures that the extracted drift component physically conforms to the aging law of the sensor and will not mistakenly absorb the impact trend of early faults into the drift baseline, thereby protecting the fault features from being "washed away".

[0010] S203. The vibration signal value of the slow-drift baseline function is used as the zero-point slow-drift component. The zero-point slow-drift component is subtracted from the original signal to obtain the raw signal. Specifically: S de-drift (t) = S raw (t)-D(t),S raw (t) represents the vibration signal in the original signal; S de-drift (t) represents the raw signal. The first stage of signal purification is completed, providing an "unbiased" data foundation for subsequent operating condition normalization and fault detection.

[0011] S300: Collects three raw signals from mining equipment during normal operation and divides them into working condition grids. Constructs a feature fingerprint vector of the vibration signal in each working condition grid. Finds the feature fingerprint vector of the corresponding vibration signal based on the current temperature and current load rate, and normalizes the current gross signal. Furthermore, the specific steps for normalizing the current hair signal are as follows: S301. Collect three original signals from the mining equipment during normal operation in the past, and divide the load rate signal and temperature signal into N working condition grids. The division rule means that the load rate signal is one grid for every 5% and the temperature signal is one grid for every 10℃. In each grid, the characteristic fingerprint of the vibration signal under the corresponding working condition is calculated, and the characteristic fingerprint vector is F. i，j =[P peak P rms P skew P kurt ], P peak P represents the peak value of the vibration signal. rms P represents the effective value of the vibration signal. skew P represents the skewness of the vibration signal. kurt The kurtosis of the vibration signal is represented; the effective value is the square root of the squared vibration signal, averaged, and then taken; F i，j It represents the grid feature fingerprint vector of the i-th horizontal axis and the j-th vertical axis; using multidimensional statistics to lock the "normal signal morphology" under this working condition is more robust than a single effective value and can capture signal waveform distortion caused by changes in working conditions.

[0012] S303. In the current moment's gross signal, extract the feature fingerprint vector that is identical to the current load rate signal and temperature signal from the operating condition grid as the current feature fingerprint vector F. currentNormalization is performed using the current feature fingerprint vector, using the following formula: ; In the formula, S normalized (t) represents the normalized pure residual signal, P rms current P represents the valid value in the current feature fingerprint vector. rms baseline This represents the median of the effective values ​​of the vibration signal across all historical operating condition grids.

[0013] By forcibly "flattening" the vibration amplitude under different operating conditions to a unified baseline, subsequent energy detection is no longer affected by load / temperature changes, and only the true fault energy will be identified.

[0014] S400: Construct a short-time energy observation window and calculate the energy of the vibration signal in the short-time energy observation window; calculate the normal energy envelope using the energy of the vibration signal during the historical normal operation of the mining equipment, calculate the residual energy ratio, and use the residual energy ratio to determine the activation of fault symptoms; Furthermore, the specific steps for determining fault symptom activation using the residual energy ratio are as follows: S401, Construct a short-time energy observation window W obs The window length is 10s. The energy of the vibration signal within the short-time energy observation window is calculated using the following formula: E obs This represents the energy of the vibration signal within the short-time energy observation window; Vibration signals from mining equipment during normal operation are collected. The energy of the short-time energy observation window vibration signal during normal operation is calculated as the normal energy. A normal energy envelope varying with operating conditions is constructed for the normal energy of different grids. The residual energy ratio is calculated using the following formula: ; In the formula, R E (t) represents the residual energy ratio at the current time, E obs (t) represents the energy of the vibration signal within the short-time energy observation window at the current moment, E normal (t) represents the normal energy corresponding to the current working condition grid; S402, Preset energy ratio threshold θ E When R E (t) > θ E Furthermore, if the energy ratio exceeds the threshold for more than M consecutive short-term energy observation windows, the fault symptom is determined to be activated.

[0015] It provides a dynamic energy reference that changes in real time with load and temperature, rather than a fixed constant, so that the residual energy ratio accurately reflects "extra energy beyond normal operating conditions", i.e., fault symptoms.

[0016] S500: Establish a symptom phase latch, use the residual energy ratio to perform iterative calculations on the symptom phase latch, and use the iteration results to determine whether the fault symptom is a valid symptom. Furthermore, the specific steps for determining whether a fault symptom is a valid symptom using the iteration results are as follows: S501. Early-stage mine faults exhibit a monotonically worsening physical characteristic. In contrast, energy fluctuations caused by random noise or external disturbances are often random and non-monotonic.

[0017] The symptom phase latch is established as Φ(t) = Φ(t-Δt) + ΔΦ(t); Φ(t) represents the symptom phase latch value, which is dimensionless; Δt represents the sliding step size of the short-time energy observation window; ΔΦ(t) represents the change value of the symptom phase latch. The residual energy ratio is used to iteratively calculate the symptom phase latch, specifically: when R E (t) > 0 and R E (t) > R E When (t-Δt), ΔΦ(t) = +1; When R E (t) < 0 and R E (t) <R E When (t-Δt), ΔΦ(t) = -k×Φ(t-Δt), where k represents the forgetting factor; Iterative calculations allow for short-term fluctuations, but only signs of continuous deterioration accumulate to a threshold that triggers an effective warning, significantly reducing the false alarm rate without sacrificing sensitivity to progressive failures.

[0018] S502. Each time a fault symptom is detected as activated, a symptom phase latch is used for calculation, with a preset cumulative threshold Φy. After iterative calculation, only when Φ(t) > Φy is the fault symptom output as a valid symptom; otherwise, the activation of the fault symptom is judged as transient interference and is masked. This forms a second "credibility gating" to ensure that the output warning has physical evolution support, rather than data randomness.

[0019] S600: Short-term and long-term baselines are constructed by using the residual energy difference of different time periods in the historical operation of mining equipment. Early warning thresholds are dynamically defined using the short-term and long-term baselines. When the early warning threshold is used to make a final judgment on the current residual energy difference when it is judged to be a valid sign, an early warning is triggered.

[0020] Furthermore, the specific steps for triggering an early warning by using an early warning threshold to make a final judgment on the current residual energy difference are as follows: S601, The short-term baseline represents the median μ of the residual energy ratio over the historical 24 hours. 24h The long-term baseline represents the median μ of the residual energy ratio over the historical 30 days.30d ; The warning threshold is dynamically calculated using short-term and long-term baselines, using the following formula: θ dynamic =max(μ 30d +β×σ 30d μ 30d +α×(μ 24h -μ 30d )+β×σ 30d ); In the formula, θ dynamic α represents the warning threshold, β represents the operating condition adaptive coefficient, and σ represents the alarm sensitivity coefficient. Both α and β are set by the operator. 30d The standard deviation of the residual energy ratio over a historical 30-day period; The maximum of the two values ​​is used to ensure that the threshold never falls below the physical safety threshold, balancing sensitivity and robustness. α allows for proactively raising the threshold when recent fluctuations intensify and proactively lowering the threshold when it becomes negative.

[0021] S602. When the fault symptom is determined to be a valid symptom, the current residual energy ratio is judged using the early warning threshold. When R E (t) > θ dynamic When this happens, an early warning is triggered, and the vibration signal in the current original signal is locked for staff to review.

[0022] A mine equipment fault early warning system based on multi-source data includes a signal segmentation module, a steady-state removal module, a normalization module, a fault symptom judgment module, a symptom confirmation module, and a final early warning module. The signal segmentation module is used to set up sensors in the mining equipment to form monitoring points. The monitoring points are used to simultaneously collect three raw signals: vibration signal, temperature signal and current load rate signal when the mining equipment is working. The signal time axis is divided into three semantic segments: steady state segment, transient segment and transition segment. The steady-state removal module is used to set a sliding freeze window, perform median fitting of the steady-state segment of the vibration signal to form a slow drift baseline function, use the vibration signal value of the slow drift baseline function as the zero-point slow drift component, and subtract the zero-point slow drift component from the original signal to obtain the gross signal. The normalization module is used to collect three original signals from the mining equipment during its historical normal operation and divide them into working condition grids. In each working condition grid, a feature fingerprint vector of the vibration signal is constructed. The feature fingerprint vector of the corresponding vibration signal is found according to the current temperature and the current load rate, and the current gross signal is normalized. The fault symptom judgment module is used to construct a short-time energy observation window, calculate the energy of the vibration signal in the short-time energy observation window, calculate the normal energy envelope using the energy of the vibration signal during the historical normal operation of the mining equipment, calculate the residual energy ratio, and use the residual energy ratio to determine the activation of the fault symptom. The symptom confirmation module is used to establish a symptom phase latch, perform iterative calculations on the symptom phase latch using the residual energy ratio, and use the iteration results to determine whether the fault symptom is a valid symptom. The final early warning module is used to construct short-term and long-term baselines by utilizing the residual energy difference of different time periods in the historical operation of mining equipment. The early warning threshold is dynamically defined using the short-term and long-term baselines. When the early warning threshold is used to make a final judgment on the current residual energy difference when it is determined to be a valid sign, an early warning is triggered.

[0023] The steady-state removal module includes function construction units and removal units; The function construction unit is used to fit the anchor trajectory of the median sequence as the sensor drift using the minimum energy path constraint to obtain the slow drift baseline function; The removal unit is used to take the slow drift baseline function vibration signal value as the zero-point slow drift component, and subtract the zero-point slow drift component from the original signal to obtain the raw signal.

[0024] Compared with the prior art, the beneficial effects of the present invention are: 1. Through five barriers—steady-state screening, drift separation, operating condition normalization, phase latch, and dynamic threshold—it effectively shields operating condition fluctuations and random transient interference. Normalization "strips" fault energy from operating condition fluctuations, and the accumulation mechanism of the phase latch captures gradual deterioration, preventing early, minor faults from being overwhelmed.

[0025] 2. The system output is not a binary "alarm / no alarm", but a "valid sign" accumulated by the phase latch, with an energy ratio amplitude. This helps maintenance personnel distinguish between "immediately required" and "observable" faults, reducing the number of unnecessary downhole inspections.

[0026] 3. Through a three-layer purification architecture of physical semantic segmentation, drift separation, and operating condition normalization, the signal source is "returned to its proper place" rather than relying on a "one-size-fits-all" approach using a general filter. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the structure of a mining equipment fault early warning method based on multi-source data according to the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] Example: Figure 1 As shown, the present invention provides a technical solution. A method for early warning of mining equipment faults based on multi-source data, the method comprising the following steps: S100. Sensors are set up in the mining equipment to form monitoring points. The monitoring points are used to synchronously collect three raw signals: vibration signal, temperature signal and current load rate signal when the mining equipment is working. The signal time axis is divided into three semantic segments: steady state segment, transient segment and transition segment. The specific steps for dividing the signal time axis into three semantic segments—steady-state segment, transient segment, and transition segment—are as follows: S101. Calculate the rate of change R of the current load rate signal of the mining equipment. L =dL(t) / dt, where L(t) represents the current load rate of the mining equipment; the vibration signal represents the vibration acceleration of the equipment; Preset noise thresholds δ1 and impulse thresholds δ2, in the signal time axis when |R L When | ≤ δ1, it is divided into a steady state segment; when δ1 < |R L When | < δ2, it is divided into a transition segment; when | R L When |≥δ2, it is divided into transient segments.

[0030] The steady-state segment indicates that the load rate change is completely submerged in sensor noise and can be regarded as zero change; the transition segment indicates that it is neither equal to noise nor equal to impact, covering common scenarios in mines such as gradual change in tunneling hardness and slow accumulation of material on conveyor belts; the transient segment indicates equipment startup, emergency shutdown, and sudden impact of the bucket against the rock wall, which will be given lower weight in subsequent steps. The three original signals provide physical "time anchors" for the separation of different signal components, avoiding confusion between normal transient behaviors such as startup and shutdown and sensor drift or equipment failure, thus reducing the false alarm rate from the source.

[0031] The introduction of a "transition section" covers scenarios such as gradual changes in tunneling hardness and slow material accumulation, filling the gap in traditional methods that only distinguish between "steady state / transient state". This allows the unique gradual working conditions of mines to be preserved and given reasonable weight, avoiding misjudgment of gradual working conditions as faults.

[0032] S200. Set a sliding freeze window, perform median fitting of the steady-state segment of the vibration signal to form a slow drift baseline function, take the vibration signal value of the slow drift baseline function as the zero-point slow drift component, and subtract the zero-point slow drift component from the original signal to obtain the gross signal. The specific steps for subtracting the zero-point slow drift component from the original signal to obtain the gross signal are as follows: S201, The length of the sliding freeze window is T. freeze The time frame can be 72 hours, with a step size of 1 hour. Within each sliding freeze window, the median M of the vibration signals in all steady-state segments is extracted. Over n consecutive sliding freeze windows, the median sequence {M1, M2, M3...M...} is generated. n The anchor point trajectories, M1, M2, M3...M, are considered as sensor drift. n This represents the median of the vibration signal for the 1st, 2nd, 3rd...nth consecutive sliding freeze window; By utilizing the physical characteristic of "extremely slow" drift, the zero-point drift that takes months to separate from the vibration signal that takes seconds or minutes, thus avoiding the damage to low-frequency fault components caused by traditional high-pass filtering.

[0033] S202. Fit the anchor point trajectory using the minimum energy path constraint to obtain the slow drift baseline function D(t); the minimum energy path constraint includes the first derivative boundedness and the second derivative smoothing, the first derivative boundedness means |dD(t) / dt|<ε, ε represents the maximum allowable drift rate, which is determined by the sensor manual; The second derivative smoothing means that the second derivative of D(t) takes a minimum value, which is set by the staff; this ensures that the extracted drift component physically conforms to the aging law of the sensor and will not mistakenly absorb the impact trend of early faults into the drift baseline, thereby protecting the fault features from being "washed away".

[0034] S203. The vibration signal value of the slow-drift baseline function is used as the zero-point slow-drift component. The zero-point slow-drift component is subtracted from the original signal to obtain the raw signal. Specifically: S de-drift (t) = S raw (t)-D(t),S raw (t) represents the vibration signal in the original signal; S de-drift (t) represents the raw signal. The first stage of signal purification is completed, providing an "unbiased" data foundation for subsequent operating condition normalization and fault detection.

[0035] S300: Collects three raw signals from mining equipment during normal operation and divides them into working condition grids. Constructs a feature fingerprint vector of the vibration signal in each working condition grid. Finds the feature fingerprint vector of the corresponding vibration signal based on the current temperature and current load rate, and normalizes the current gross signal. The specific steps for normalizing the current hair signal are as follows: S301. Collect three original signals from the mining equipment during normal operation in the past, and divide the load rate signal and temperature signal into N working condition grids. The division rule means that the load rate signal is one grid for every 5% and the temperature signal is one grid for every 10℃. In each grid, the characteristic fingerprint of the vibration signal under the corresponding working condition is calculated, and the characteristic fingerprint vector is F. i，j =[P peak P rms P skew P kurt ], P peak P represents the peak value of the vibration signal. rms P represents the effective value of the vibration signal. skew P represents the skewness of the vibration signal. kurt The kurtosis of the vibration signal is represented; the effective value is the square root of the squared vibration signal, averaged, and then taken; F i，j It represents the grid feature fingerprint vector of the i-th horizontal axis and the j-th vertical axis; using multidimensional statistics to lock the "normal signal morphology" under this working condition is more robust than a single effective value and can capture signal waveform distortion caused by changes in working conditions.

[0036] S303. In the current moment's gross signal, extract the feature fingerprint vector that is identical to the current load rate signal and temperature signal from the operating condition grid as the current feature fingerprint vector F. current Normalization is performed using the current feature fingerprint vector, using the following formula: ; In the formula, S normalized (t) represents the normalized pure residual signal, P rms current P represents the valid value in the current feature fingerprint vector. rms baseline This represents the median of the effective values ​​of the vibration signal across all historical operating condition grids.

[0037] By forcibly "flattening" the vibration amplitude under different operating conditions to a unified baseline, subsequent energy detection is no longer affected by load / temperature changes, and only the true fault energy will be identified.

[0038] S400: Construct a short-time energy observation window and calculate the energy of the vibration signal in the short-time energy observation window; calculate the normal energy envelope using the energy of the vibration signal during the historical normal operation of the mining equipment, calculate the residual energy ratio, and use the residual energy ratio to determine the activation of fault symptoms; The specific steps for determining fault symptom activation using residual energy ratio are as follows: S401, Construct a short-time energy observation window W obsThe window length is 10s. The energy of the vibration signal within the short-time energy observation window is calculated using the following formula: E obs This represents the energy of the vibration signal within the short-time energy observation window; Vibration signals from mining equipment during normal operation are collected. The energy of the short-time energy observation window vibration signal during normal operation is calculated as the normal energy. A normal energy envelope varying with operating conditions is constructed for the normal energy of different grids. The residual energy ratio is calculated using the following formula: ; In the formula, R E (t) represents the residual energy ratio at the current time, E obs (t) represents the energy of the vibration signal within the short-time energy observation window at the current moment, E normal (t) represents the normal energy corresponding to the current working condition grid; S402, Preset energy ratio threshold θ E When R E (t) > θ E Furthermore, if the energy ratio exceeds the threshold for more than M consecutive short-term energy observation windows, the fault symptom is determined to be activated.

[0039] It provides a dynamic energy reference that changes in real time with load and temperature, rather than a fixed constant, so that the residual energy ratio accurately reflects "extra energy beyond normal operating conditions", i.e., fault symptoms.

[0040] S500: Establish a symptom phase latch, use the residual energy ratio to perform iterative calculations on the symptom phase latch, and use the iteration results to determine whether the fault symptom is a valid symptom. The specific steps for determining whether a fault symptom is a valid symptom using the iteration results are as follows: S501. Early-stage mine faults exhibit a monotonically worsening physical characteristic. In contrast, energy fluctuations caused by random noise or external disturbances are often random and non-monotonic.

[0041] The symptom phase latch is established as Φ(t) = Φ(t-Δt) + ΔΦ(t); Φ(t) represents the symptom phase latch value, which is dimensionless; Δt represents the sliding step size of the short-time energy observation window; ΔΦ(t) represents the change value of the symptom phase latch. The residual energy ratio is used to iteratively calculate the symptom phase latch, specifically: when R E (t) > 0 and R E (t) > R E When (t-Δt), ΔΦ(t) = +1; When R E (t) < 0 and R E (t) <R EWhen (t-Δt), ΔΦ(t) = -k×Φ(t-Δt), where k represents the forgetting factor; Iterative calculations allow for short-term fluctuations, but only signs of continuous deterioration accumulate to a threshold that triggers an effective warning, significantly reducing the false alarm rate without sacrificing sensitivity to progressive failures.

[0042] S502. Each time a fault symptom is detected as activated, a symptom phase latch is used for calculation, with a preset cumulative threshold Φy. After iterative calculation, only when Φ(t) > Φy is the fault symptom output as a valid symptom; otherwise, the activation of the fault symptom is judged as transient interference and is masked. This forms a second "credibility gating" to ensure that the output warning has physical evolution support, rather than data randomness.

[0043] S600: Short-term and long-term baselines are constructed by using the residual energy difference of different time periods in the historical operation of mining equipment. Early warning thresholds are dynamically defined using the short-term and long-term baselines. When the early warning threshold is used to make a final judgment on the current residual energy difference when it is judged to be a valid sign, an early warning is triggered.

[0044] The specific steps for triggering an early warning by using an early warning threshold to make a final judgment on the current residual energy difference are as follows: S601, The short-term baseline represents the median μ of the residual energy ratio over the historical 24 hours. 24h The long-term baseline represents the median μ of the residual energy ratio over the historical 30 days. 30d ; The warning threshold is dynamically calculated using short-term and long-term baselines, using the following formula: θ dynamic =max(μ 30d +β×σ 30d μ 30d +α×(μ 24h -μ 30d )+β×σ 30d ); In the formula, θ dynamic α represents the warning threshold, β represents the operating condition adaptive coefficient, and σ represents the alarm sensitivity coefficient. Both α and β are set by the operator. 30d The standard deviation of the residual energy ratio over a historical 30-day period; The maximum of the two values ​​is used to ensure that the threshold never falls below the physical safety threshold, balancing sensitivity and robustness. α allows for proactively raising the threshold when recent fluctuations intensify and proactively lowering the threshold when it becomes negative.

[0045] S602. When the fault symptom is determined to be a valid symptom, the current residual energy ratio is judged using the early warning threshold. When R E (t) > θ dynamicWhen this happens, an early warning is triggered, and the vibration signal in the current original signal is locked for staff to review.

[0046] A mine equipment fault early warning system based on multi-source data includes a signal segmentation module, a steady-state removal module, a normalization module, a fault symptom judgment module, a symptom confirmation module, and a final early warning module. The signal segmentation module is used to set up sensors in the mining equipment to form monitoring points. The monitoring points are used to simultaneously collect three raw signals: vibration signal, temperature signal and current load rate signal when the mining equipment is working. The signal time axis is divided into three semantic segments: steady state segment, transient segment and transition segment. The steady-state removal module is used to set a sliding freeze window, perform median fitting of the steady-state segment of the vibration signal to form a slow drift baseline function, use the vibration signal value of the slow drift baseline function as the zero-point slow drift component, and subtract the zero-point slow drift component from the original signal to obtain the gross signal. The normalization module is used to collect three original signals from the mining equipment during its historical normal operation and divide them into working condition grids. In each working condition grid, a feature fingerprint vector of the vibration signal is constructed. The feature fingerprint vector of the corresponding vibration signal is found according to the current temperature and the current load rate, and the current gross signal is normalized. The fault symptom judgment module is used to construct a short-time energy observation window, calculate the energy of the vibration signal in the short-time energy observation window, calculate the normal energy envelope using the energy of the vibration signal during the historical normal operation of the mining equipment, calculate the residual energy ratio, and use the residual energy ratio to determine the activation of the fault symptom. The symptom confirmation module is used to establish a symptom phase latch, perform iterative calculations on the symptom phase latch using the residual energy ratio, and use the iteration results to determine whether the fault symptom is a valid symptom. The final early warning module is used to construct short-term and long-term baselines by utilizing the residual energy difference of different time periods in the historical operation of mining equipment. The early warning threshold is dynamically defined using the short-term and long-term baselines. When the early warning threshold is used to make a final judgment on the current residual energy difference when it is determined to be a valid sign, an early warning is triggered.

[0047] The steady-state removal module includes function construction units and removal units; The function construction unit is used to treat the median sequence as the anchor trajectory of sensor drift and fit the anchor trajectory using the minimum energy path constraint to obtain the slow drift baseline function; The removal unit is used to take the slow drift baseline function vibration signal value as the zero-point slow drift component, and subtract the zero-point slow drift component from the original signal to obtain the raw signal.

[0048] Example: The drive reducer of the underground scraper conveyor was selected as the early warning target. This equipment operates under variable load and temperature for a long time, and the sensor drifts to zero point for a long time. There is a lot of underground impact interference, and the false alarm rate of traditional fixed threshold early warning is high.

[0049] Set the energy ratio threshold to 1.3, continuous window M to 5, forgetting factor to 0.1, cumulative threshold to 11, adaptive coefficient to 0.6, and sensitivity coefficient to 1.8. The gateway synchronously acquires three signals: vibration acceleration of 0.82g, oil temperature of 55℃, real-time load rate of 45%, and load change rate of 0.2% / s, and outputs one set of time-series data per second; the original signal is divided into three segments. Construct a historical fitting drift baseline D(t) = 0.06g. Current operating condition mesh (load 45%, oil temperature 55℃) characteristic fingerprint: F current =[2.1, 0.42, 0.12, 3.15]; The median effective value of the vibration signal in all historical working condition grids is 0.3, and the normal short-time energy of the current working condition is calculated to be 0.152g. 2 •s; short-term baseline is 1.55, long-term baseline is 1.1, and standard deviation is 0.22; Subtracting the slow zero-point drift component from the original signal yields the gross signal = 0.82 − 0.06 = 0.76g; After normalization, the result is 1.064g. The energy of the current window, calculated by on-site integration, is 0.248g. 2 ·s, calculated residual energy ratio = 1.6316; The fault symptom is activated when the current window is greater than the energy ratio threshold and this condition is met for five consecutive sliding windows. If the residual energy ratio at the previous time step is 1.51, and the current time step is greater than 0 and greater than the previous time step, increment the latch value by one. Assuming the latch value was 11 in the previous moment, and the current accumulated value is 12, which is greater than the cumulative threshold, it is determined to be a valid fault symptom; the warning threshold is calculated to be 1.766; the current residual energy ratio is less than the warning threshold, so no formal warning is triggered this time.

[0050] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for early warning of mining equipment faults based on multi-source data, characterized in that: The method includes the following steps: S100. Sensors are set up in the mining equipment to form monitoring points. The monitoring points are used to synchronously collect three raw signals: vibration signal, temperature signal and current load rate signal when the mining equipment is working. The signal time axis is divided into three semantic segments: steady state segment, transient segment and transition segment. S200. Set a sliding freeze window, perform median fitting of the steady-state segment of the vibration signal to form a slow drift baseline function, take the vibration signal value of the slow drift baseline function as the zero-point slow drift component, and subtract the zero-point slow drift component from the original signal to obtain the gross signal. S300: Collects three raw signals from mining equipment during normal operation and divides them into working condition grids. Constructs a feature fingerprint vector of the vibration signal in each working condition grid. Finds the feature fingerprint vector of the corresponding vibration signal based on the current temperature and current load rate, and normalizes the current gross signal. S400: Construct a short-time energy observation window and calculate the energy of the vibration signal in the short-time energy observation window; calculate the normal energy envelope using the energy of the vibration signal during the historical normal operation of the mining equipment, calculate the residual energy ratio, and use the residual energy ratio to determine the activation of fault symptoms; S500: Establish a symptom phase latch, use the residual energy ratio to perform iterative calculations on the symptom phase latch, and use the iteration results to determine whether the fault symptom is a valid symptom. S600: Short-term and long-term baselines are constructed by using the residual energy difference of different time periods in the historical operation of mining equipment. Early warning thresholds are dynamically defined using the short-term and long-term baselines. When the early warning threshold is used to make a final judgment on the current residual energy difference when it is judged to be a valid sign, an early warning is triggered.

2. The method for early warning of mining equipment faults based on multi-source data according to claim 1, characterized in that: In S100, the signal time axis is divided into three semantic segments: steady-state segment, transient segment, and transition segment. Specifically: Calculate the rate of change R of the current load rate signal of the mining equipment. L =dL(t) / dt, where L(t) represents the current load rate of the mining equipment; the vibration signal represents the vibration acceleration of the equipment; Preset noise thresholds δ1 and impulse thresholds δ2, in the signal time axis when |R L When | ≤ δ1, it is divided into a steady state segment; when δ1 < |R L When | < δ2, it is divided into a transition segment; when | R L When |≥δ2, it is divided into transient segments.

3. The method for early warning of mining equipment faults based on multi-source data according to claim 1, characterized in that: The slow-drift baseline function constructed by fitting the median of the steady-state segment of the vibration signal in S200 is specifically as follows: The length of the sliding freeze window is T. freeze In each sliding freeze window, the median M of the vibration signals within all steady-state segments is extracted. Over n consecutive sliding freeze windows, the median sequence {M1, M2, M3...M...} is calculated. n The anchor point trajectories, M1, M2, M3...M, are considered as sensor drift. n This represents the median of the vibration signal for the 1st, 2nd, 3rd...nth consecutive sliding freeze window; The anchor point trajectory is fitted using the minimum energy path constraint to obtain the slow drift baseline function D(t); The minimum energy path constraint includes first-order derivative boundedness and second-order derivative smoothing. The first-order derivative boundedness means |dD(t) / dt|<ε, where ε represents the maximum allowable drift rate, which is determined by the sensor manual. The second derivative smoothing means that the second derivative of D(t) takes a minimum value, which is set by the staff.

4. The method for early warning of mining equipment faults based on multi-source data according to claim 1, characterized in that: Specifically, in S200, subtracting the zero-point slow drift component from the original signal to obtain the gross signal is as follows: The vibration signal value of the slow-drift baseline function is taken as the zero-point slow-drift component. The zero-point slow-drift component is subtracted from the original signal to obtain the raw signal, specifically: S de-drift (t) = S raw (t)-D(t),S raw (t) represents the vibration signal in the original signal; S de-drift (t) represents the hair signal.

5. The method for early warning of mining equipment faults based on multi-source data according to claim 1, characterized in that: The normalization of the current hair signal in S300 specifically involves: Three raw signals from the mining equipment during its historical normal operation were collected, and N working condition grids were divided using the load rate signal and temperature signal. In each grid, the characteristic fingerprint of the vibration signal under the corresponding working condition is calculated, and the characteristic fingerprint vector is F. i，j =[P peak P rms P skew P kurt ], P peak P represents the peak value of the vibration signal. rms P represents the effective value of the vibration signal. skew P represents the skewness of the vibration signal. kurt The kurtosis of the vibration signal is represented; the effective value is the square root of the squared vibration signal, averaged, and then taken; F i，j This represents the grid feature fingerprint vector of the i-th horizontal axis and the j-th vertical axis; In the current moment's gross signal, extract the feature fingerprint vector that is identical to the current load rate signal and temperature signal from the load cell grid as the current feature fingerprint vector F. current The clean residual signal is obtained by normalizing the current feature fingerprint vector.

6. The method for early warning of mining equipment faults based on multi-source data according to claim 1, characterized in that: The specific method for determining fault symptom activation using residual energy ratio in S400 is as follows: Constructing a short-time energy observation window W obs The energy E of the vibration signal within the short-time energy observation window is calculated by integrating the pure residual signal. obs ; Vibration signals from mining equipment during normal operation are collected. The energy of the short-time energy observation window vibration signal during normal operation is calculated as the normal energy. The normal energy envelope of the grid under different working conditions is constructed to vary with the working conditions. The formula for calculating the residual energy ratio is: ； In the formula, R E (t) represents the residual energy ratio at the current time, E obs (t) represents the energy of the vibration signal within the short-time energy observation window at the current moment, E normal (t) represents the normal energy corresponding to the current working condition grid; S402, Preset energy ratio threshold θ E When R E (t) > θ E Furthermore, if the energy ratio exceeds the threshold for more than M consecutive short-term energy observation windows, the fault symptom is determined to be activated.

7. The method for early warning of mining equipment faults based on multi-source data according to claim 1, characterized in that: In S500, the specific steps for determining whether a fault symptom is a valid symptom using the iteration results are as follows: The symptom phase latch is established as Φ(t) = Φ(t-Δt) + ΔΦ(t); Φ(t) represents the symptom phase latch value, which is dimensionless; Δt represents the sliding step size of the short-time energy observation window; ΔΦ(t) represents the change value of the symptom phase latch. The residual energy ratio is used to iteratively calculate the symptom phase latch, specifically: when R E (t) > 0 and R E (t) > R E When (t-Δt), ΔΦ(t) = +1; When R E (t) < 0 and R E (t) <R E When (t-Δt), ΔΦ(t) = -k×Φ(t-Δt), where k represents the forgetting factor; Each time a fault symptom is detected as active, a symptom phase latch is used for calculation, with a preset cumulative threshold Φy. After iterative calculation, when Φ(t) > Φy, the fault symptom is output as a valid symptom; otherwise, the activation of the fault symptom is judged as transient interference and is shielded.

8. The method for early warning of mining equipment faults based on multi-source data according to claim 1, characterized in that: The final determination to trigger the warning in S600 is specifically as follows: The short-term baseline represents the median μ of the residual energy ratio over the historical 24-hour period. 24h The long-term baseline represents the median μ of the residual energy ratio over the historical 30 days. 30d ; The warning threshold θ is dynamically calculated using short-term and long-term baselines. dynamic When a fault symptom is determined to be a valid symptom, the current residual energy ratio is judged using an early warning threshold. When R... E (t) > θ dynamic When this happens, an early warning is triggered, and the vibration signal in the current original signal is locked for staff to review.

9. A mine equipment fault early warning system based on multi-source data, characterized in that: The mine equipment fault early warning system includes a signal segmentation module, a steady-state removal module, a normalization module, a fault symptom judgment module, a symptom confirmation module, and a final early warning module. The signal segmentation module is used to set up sensors in the mining equipment to form monitoring points. The monitoring points are used to simultaneously collect three raw signals: vibration signal, temperature signal and current load rate signal when the mining equipment is working. The signal time axis is divided into three semantic segments: steady state segment, transient segment and transition segment. The steady-state removal module is used to set a sliding freeze window, perform median fitting of the steady-state segment of the vibration signal to form a slow drift baseline function, use the vibration signal value of the slow drift baseline function as the zero-point slow drift component, and subtract the zero-point slow drift component from the original signal to obtain the gross signal. The normalization module is used to collect three original signals from the mining equipment during its historical normal operation and divide them into working condition grids. In each working condition grid, a feature fingerprint vector of the vibration signal is constructed. The feature fingerprint vector of the corresponding vibration signal is found according to the current temperature and the current load rate, and the current gross signal is normalized. The fault symptom judgment module is used to construct a short-time energy observation window, calculate the energy of the vibration signal in the short-time energy observation window, calculate the normal energy envelope using the energy of the vibration signal during the historical normal operation of the mining equipment, calculate the residual energy ratio, and use the residual energy ratio to determine the activation of the fault symptom. The symptom confirmation module is used to establish a symptom phase latch, perform iterative calculations on the symptom phase latch using the residual energy ratio, and use the iteration results to determine whether the fault symptom is a valid symptom. The final early warning module is used to construct short-term and long-term baselines by utilizing the residual energy difference of different time periods in the historical operation of mining equipment. The early warning threshold is dynamically defined using the short-term and long-term baselines. When the early warning threshold is used to make a final judgment on the current residual energy difference when it is determined to be a valid sign, an early warning is triggered.

10. A mine equipment fault early warning system based on multi-source data according to claim 9, characterized in that: The steady-state removal module includes a function construction unit and a removal unit; The function construction unit is used to fit the anchor trajectory of the median sequence as the sensor drift using the minimum energy path constraint to obtain the slow drift baseline function; The removal unit is used to take the slow drift baseline function vibration signal value as the zero-point slow drift component, and subtract the zero-point slow drift component from the original signal to obtain the raw signal.

Images