Data-driven mine hoist main shaft device fault detection method and system

Abstract

The application discloses a mine hoist main shaft device fault detection method and system based on data driving, relates to the technical field of mine hoist data processing, and extracts a decentralized sequence of a vibration signal, scales output two-dimensional vector matrices at both ends by using a robust scale factor, solves an inter-end linear mapping matrix, calculates a residual error, generates a contact migration index, calculates a continuous wedge weight in combination with an index historical mean value, performs multiplication suppression on an inter-end consistency component, generates a stable input data matrix of stripped interference, calculates an end energy difference value, inputs the stable data into a convolutional neural network to generate a fault probability distribution, calculates and compares the priority of rope groove rope arrangement inspection and bearing main shaft maintenance by using bias intensity, wedge weight and fault probability, and outputs a final operation and maintenance instruction type. The application effectively decouples load mutation interference caused by a rope groove contact event, and avoids systematic misjudgment.

Description

Technical Field

[0001] This invention relates to the field of data processing technology for mine hoists, and specifically to a data-driven method and system for fault detection of the main shaft device of a mine hoist. Background Technology

[0002] As a core piece of equipment in mine production, the mine hoist's main shaft assembly operates under multi-layer winding conditions. The wire rope needs to alternately wind in and out at different diameter layers, frequently experiencing inter-coil and inter-layer transitions. In the harsh environment of long-term low-speed, heavy-load operation, severe and continuous friction occurs in the drum grooves, especially near the transition zone, easily leading to uneven groove wear. This results in steps, chamfer collapse, or localized metal burrs on the groove lip. This progressive morphological defect induces a high-probability but low-frequency abnormal phenomenon: semi-wedge-locking contact. When the wire rope is subjected to instantaneous uneven load or lubrication fluctuations in the transition zone, the rope will experience localized biting, wedging, and squeezing slippage at the lip burrs. At this moment, the conventional load transmission path of the main shaft assembly is instantly rewritten, abruptly changing from primarily transmitting torque and stabilizing radial loads to a superimposed short-term lateral thrust and additional bending moment. This drastic change in the force transmission structure will induce strong and spatially asymmetrical abnormal vibration responses on the bearing seats at both ends of the main shaft.

[0003] Current mainstream fault diagnosis systems have significant limitations when handling the aforementioned complex operating conditions. Current diagnostic schemes typically assume statically that all acquired abnormal impacts originate from localized damage to the bearing or spindle itself, and often construct state discrimination models based on standard fundamental fault modes. However, when a semi-wedge-in locking rope contact event occurs, the multi-channel vibration signals acquired by the system inevitably become contaminated with strong non-stationary and asymmetric structural impact interference caused by the instantaneous rewriting of the load path. If the raw vibration data, without decoupling and purification, is directly input into existing classification diagnostic models, under the dual influence of strong background noise and complex transmission constraints, the algorithm model is prone to interpretative adsorption effects. This means that the impact energy caused by the abrupt change in spatial force state is incorrectly projected and adsorbed into the closest bearing damage mode. This leads to a serious systemic misjudgment blind spot; existing diagnostic systems are highly susceptible to misjudging contact events caused by uneven wear of the drum rope groove as internal bearing damage or a serious spindle fault, thus issuing completely erroneous maintenance decision instructions to the terminal. The existing system misled the site into carrying out destructive overhauls and bearing disassembly, instead of performing low-cost rope groove repairs and rope alignment checks. This not only resulted in extremely high costs for ineffective downtime but also masked the actual operational risks and hidden dangers of the equipment. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a data-driven fault detection method and system for the main shaft device of mine hoists. This solves the problem that existing diagnostic models are prone to interpretive adsorption effects, misjudging load mutations and asymmetric impact interference caused by contact events such as rope groove wear as internal bearing damage modes, thus leading to systemic misjudgments and ineffective shutdowns.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] The decentralized sequence of the mechanical vibration signal is extracted, the median absolute deviation of the decentralized sequence is calculated to generate a robust scaling factor, and the robust scaling factor is used to scale and combine the output two-dimensional vector matrix of the driving end and the two-dimensional vector matrix of the non-driving end.

[0007] Based on the two-dimensional vector matrix of the driving end and the two-dimensional vector matrix of the non-driving end, the linear mapping matrix between the ends is solved, and the minimum mean square residual is calculated to generate the contact migration index.

[0008] The arithmetic mean of the contact migration index is calculated and the continuous wedge weight is calculated in combination with the contact migration index. Based on the inter-end linear mapping matrix and the driving end two-dimensional vector matrix, the inter-end consistency two-dimensional stable component matrix is ​​predicted and generated. The stable input data matrix is ​​generated by the continuous wedge weight multiplication suppression.

[0009] The end energy difference is calculated based on the two-dimensional vector matrix of the driving end and the two-dimensional vector matrix of the non-driving end. The stable input data matrix is ​​then input into a one-dimensional convolutional neural network to generate a fault category probability distribution array.

[0010] The bias strength is calculated using the end energy difference. The priority of rope groove rope arrangement inspection and bearing spindle maintenance is calculated by combining the continuous wedge weight, bias strength and fault category probability distribution array, and the operation and maintenance instruction type is output.

[0011] Furthermore, a data-driven fault detection system for the main shaft assembly of a mine hoist is proposed to implement any of the above-mentioned data-driven fault detection methods for the main shaft assembly of a mine hoist, including:

[0012] The data acquisition module is used to extract the decentralized sequence of mechanical vibration signals, calculate the median absolute deviation of the decentralized sequence to generate a robust scaling factor, and use the robust scaling factor to scale and combine the output two-dimensional vector matrix of the driving end and the two-dimensional vector matrix of the non-driving end.

[0013] The wedge characterization module is used to solve the inter-end linear mapping matrix based on the two-dimensional vector matrix of the driving end and the two-dimensional vector matrix of the non-driving end, and to calculate the minimum mean square residual to generate the contact migration index.

[0014] The load decoupling module is used to calculate the arithmetic mean of the contact migration index and combine it with the contact migration index to calculate the continuous wedge weight. Based on the inter-end linear mapping matrix and the two-dimensional vector matrix of the driving end, it predicts and generates an inter-end consistent two-dimensional stable component matrix. After continuous wedge weight multiplication suppression, a stable input data matrix is ​​generated.

[0015] The state discrimination module is used to calculate the end energy difference based on the two-dimensional vector matrix of the driving end and the two-dimensional vector matrix of the non-driving end, and input the stable input data matrix into a one-dimensional convolutional neural network to generate a fault category probability distribution array.

[0016] The instruction generation module is used to calculate the bias strength using the end energy difference, and to calculate the rope groove rope arrangement inspection priority and bearing spindle maintenance priority by combining the continuous wedge weight, bias strength and fault category probability distribution array, and output the operation and maintenance instruction type.

[0017] Compared with existing technologies, it has the following advantages:

[0018] This paper proposes a data-driven fault detection method and system for mine hoist main shaft devices. Addressing the problem that traditional diagnostic models easily misinterpret asymmetric impact energy caused by rope groove wear as bearing damage modes, a load decoupling and clean data extraction mechanism is constructed. By solving the inter-end linear mapping matrix and contact migration index, the system dynamically quantifies the deviation of the spatial load transmission path. Relying on a continuously updated historical background queue to generate continuous wedge weights, the system performs multiplicative suppression on the consistency components at both ends, substantially removing non-stationary impact interference caused by rope groove contact behavior from the main data stream. This mechanism abandons the rigid hard threshold judgment of existing technologies, avoids the continuous contamination of the classification model by structural asymmetric vibration, ensures the purity of the data input to the convolutional neural network, and effectively eliminates the blind spot of systematic misjudgment caused by sudden load changes.

[0019] Based on the acquisition of pure vibration data, this solution further establishes a dual-line dynamic competition architecture for maintenance commands. The system calculates the energy difference in the horizontal and vertical directions to form the bias intensity, and combines the output probability of a convolutional neural network to independently quantify the priority of rope groove inspection and bearing spindle maintenance. Through the comparison and adjudication of the dual priorities, the system successfully transforms the obscure vibration analysis results into accurate execution commands targeting specific structural components. When load path rewriting dominates, the system outputs a rope groove inspection command and provides end-direction bias clues; when internal damage dominates, it outputs a maintenance command. This adjudication mechanism resolves the long-standing industry dilemma of blindly disassembling the spindle due to rope groove wear, significantly reduces the ineffective downtime costs caused by diagnostic errors, and improves the reliability and safety of the hoist's daily operation. Attached Figure Description

[0020] Figure 1This is a schematic diagram of the system framework of the present invention.

[0021] Figure 2 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0022] 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.

[0023] Please see Figures 1 to 2 This application provides a data-driven method for fault detection of the main shaft device of a mine hoist;

[0024] The method specifically includes the following steps:

[0025] Step 1: During multi-layer winding operation, the main shaft assembly of a mine hoist is susceptible to interference from the semi-wedge-locking rope contact phenomenon. This interference can cause asymmetrical strong impacts on the bearing seats at both ends of the main shaft in both horizontal and vertical directions. To accurately identify the load path rewriting phenomenon in subsequent processing, Step 1 aims to construct a four-channel two-dimensional vector sequence that is resistant to accidental impacts and has uniform scale. The specific operation is as follows:

[0026] The system synchronously acquires mechanical vibration signals throughout its entire lifespan using accelerometers fixedly mounted on the bearing seats at both ends of the spindle assembly, and discretizes and segments them in the time dimension. The system's preset sampling frequency is sf (in Hertz). The sliding window duration is wt (in seconds). The window step size is st (in seconds). Based on the above global configuration parameters, the system first calculates the total number of sample points np within a single sliding window, using the following formula:

[0027] np = sf × wt;

[0028] The above formula indicates that the total number of sample points within the sliding window is equal to the product of the sampling frequency and the sliding window duration. Specifically, within any current sliding window, the system synchronously captures and outputs four-channel raw acceleration sequences. The time index of the data points in each sequence is denoted as ti, where the value of ti ranges from 1 to np. The four raw input sequences are respectively labeled as the driving end horizontal raw acceleration sequence ax[ti], the driving end vertical raw acceleration sequence ay[ti], the non-driving end horizontal raw acceleration sequence bx[ti], and the non-driving end vertical raw acceleration sequence by[ti].

[0029] Specifically, the accelerometer and data acquisition card are fundamental hardware devices for acquiring the second derivative information of mechanical vibration displacement. The sampling frequency sf determines the data resolution for capturing high-frequency transient impact characteristics, while the sliding window duration wt and window step duration st, as time scale parameters, jointly control the frequency of data stream interception and overlapping updates. High-frequency multi-channel synchronous acquisition based on the same time base ensures strict phase alignment of the four-channel data within the same sliding window. This strict phase alignment is a necessary prerequisite for subsequent steps to capture the inter-end asymmetry caused by the instantaneous rewriting of the load path.

[0030] It should be noted that, in this process, the sampling frequency sf is preferably between 25600 and 51200, and in this embodiment, it is fixed at 38580. The sliding window duration wt is preferably 2, and the window step duration st is preferably 1, i.e., the window overlap rate is 50%. The above-mentioned preferred parameters ensure that the sliding window can completely cover the half-cycle to full-cycle dynamic evolution process of the semi-wedge-locked rope contact phenomenon, not only completely encompassing the entire process of the event, but also ensuring that the statistics have sufficient sample support. The essence of the discretization operation is to map the continuous-time signal into a digital space array with fixed dimensions in computer memory, defining a clear data boundary for subsequent matrix operations.

[0031] The system performs outlier repair on the sequences ax, ay, bx, and by. If invalid values ​​(non-numerical or infinite) exist in the sequence, they are replaced with the median of the valid values ​​in the corresponding channel within the window. If there is a saturation overflow value exceeding the upper limit of the acquisition card, it is forcibly truncated to the full-scale upper limit. Subsequently, a decentralization operation, i.e., signal zeroing preprocessing, is performed on the repaired sequences. Taking the horizontal channel at the driver end as an example, the median of the window of sequence ax is calculated and subtracted from each element in the sequence to obtain the driver-end horizontal decentralized sequence cx[ti]. The same operation is performed on the other channels to generate the driver-end vertical decentralized sequence cy[ti], the non-driver-end horizontal decentralized sequence dx[ti], and the non-driver-end vertical decentralized sequence dy[ti]. The calculation formula is taken as cx[ti] as an example:

[0032]

[0033] The above formula indicates that the decentralized sequence value of a specific channel at the current time index is equal to the original acceleration sequence value of the corresponding channel at the corresponding time index, minus the median of all original acceleration sequence values ​​of the corresponding channel within the entire sliding window. The median, rather than the arithmetic mean, is used as the benchmark for decentralization because the arithmetic mean is easily skewed by a single, extremely large amplitude abrasion impact; using the median ensures a highly robust zero-point baseline.

[0034] Specifically, data contamination in industrial settings is often caused by sensor disconnections or strong electromagnetic interference. Outlier truncation and median-based replacement based on valid values ​​aim to prevent numerical divergence or collapse in subsequent inter-end mapping matrix operations. Decentralized operations utilize the median calculation mechanism from statistics to shift the signal's oscillation center to the zero mark, effectively eliminating DC static offset caused by sensor installation tilt and gravity components. This process allows the processing object to focus purely on the dynamic vibration changes of the mechanical structure, effectively eliminating the systematic impact of static bias on subsequent residual fitting.

[0035] It should be noted that the full-scale upper limit is the maximum absolute value threshold corresponding to the sensor's dynamic range, which is pre-stored in the system memory.

[0036] Calculate the median absolute deviation for each channel. Taking the driver as an example, calculate the horizontal median absolute deviation ux and the vertical median absolute deviation uy of the driver:

[0037]

[0038] Wherein, the absolute deviation of the median at the driving end level, ux, is equal to the remedian of the set of absolute values ​​of each element of the driving end level decentralized sequence minus the total median of the sequence.

[0039] The arithmetic mean of the deviations in the two directions is taken to obtain the mean value of the driver end component deviation ud, which represents the overall vibration background level of the driver end:

[0040]

[0041] Among them, the mean deviation of the driving end component ud is equal to half of the sum of the absolute deviation of the horizontal median and the absolute deviation of the vertical median of the driving end.

[0042] By introducing a minimal positive constant ep to prevent division by zero, the final robust scaling factor us at the driver end is calculated:

[0043]

[0044] Among them, the final robust scaling factor us of the driving end is equal to the square root of the sum of the mean deviation of the driving end components and the minimum positive constant.

[0045] For the non-driven end, the absolute deviation of the horizontal median vx, the absolute deviation of the vertical median vy, the mean of the component deviations vn, and the robust scaling factor vs are calculated sequentially using the sequences dx[ti] and dy[ti].

[0046]

[0047] Specifically, traditional vibration signal normalization uses standard deviation calculated based on variance. However, variance calculation amplifies extreme values ​​that deviate from the center by a quadratic order. For the short-duration, high-energy frictional impact unique to mine hoists, the traditional standard deviation is instantly amplified, resulting in severe compression of the normal vibration waveform after normalization. This solution designs a two-layer dispersion extraction mechanism centered on the median. The median has strong anti-interference capabilities against a few extreme outliers, firmly anchoring the basic energy scale of the normal operating cycle. This robust scaling factor ensures that the high-amplitude, short-duration impact characteristics generated by the semi-wedge-entry rope contact are not smoothed out but are instead fully preserved as significant features, creating a strong synergistic effect with the subsequent contact migration index extraction process.

[0048] It should be noted that the minimal positive constant ep is a necessary low-level error-proofing mechanism to ensure that the denominator still has a small non-zero value when the spindle is in a very low-speed creeping or stopped and stationary state, thereby ensuring the robustness of the system at all times. To avoid numerical interference with the normal amplitude, the magnitude of the minimal positive constant ep is limited to... to In this embodiment, the preferred embodiment is... .

[0049] Using the final robust scaling factor us at the driver end, the decentralized sequence at the driver end is divided and scaled to generate the horizontally normalized sequence px[ti] and the vertically normalized sequence py[ti] at the driver end:

[0050]

[0051] Using the final robust scaling factor vs of the non-driving end, generate the horizontally normalized sequence qx[ti] and the vertically normalized sequence qy[ti] of the non-driving end:

[0052]

[0053] Finally, the two directional components under the same time index ti are spatially combined. The driving end two-dimensional vector matrix va is formed by combining px[ti] and py[ti] point by point. The non-driving end two-dimensional vector matrix vb is formed by combining qx[ti] and qy[ti] point by point.

[0054] Specifically, scaling the decentralized pure dynamic vibration components based on the robust background energy scale of each end can substantially eliminate the inherent static differences between ends caused by asymmetric mechanical impedance or factory sensitivity errors of the sensors on both sides. Vector combination operations reduce the dimensionality of independent one-dimensional time series to a spatial vector sequence describing the motion trajectory of the centroid of a single bearing housing plane. This operation ensures rigorous comparability between the data from the driving and non-driving ends under a unified shock resistance scale, removing the dual obstacles of dimension and scale for solving the mapping relationship between ends.

[0055] It should be noted that the calculation logic of the above normalization formula is as follows: the normalized sequence value in a specific direction is equal to the decentralized sequence value in the corresponding direction divided by the final robust scaling factor of the end. Furthermore, the aforementioned driving-end two-dimensional vector matrix va and the non-driving-end two-dimensional vector matrix vb are the effective data payloads exposed to the outer layer in step one. These two matrices will serve as dedicated and direct input sources, seamlessly passed to the next level of computation specifically for constructing matrix arrays and solving for the inter-end linear mapping residuals. Specifically, in computer memory allocation, both the driving-end two-dimensional vector matrix va and the non-driving-end two-dimensional vector matrix vb are explicitly instantiated with dimensions of [dimension size missing]. A two-dimensional floating-point matrix array, where the first dimension of the matrix represents the spatial direction and the second dimension represents the time index ti. This parameter handover mechanism not only eliminates the ambiguity of disorder in the concept of mathematical sets, but also fully guarantees the closedness of the computational process and the continuity of data flow in this invention.

[0056] Step Two: When the load path of the mine hoist main shaft device is rewritten, the vibration transmission law between the ends will be disrupted. To accurately quantify the degree of this disruption, Step Two will calculate the transmission residual between the ends using the least squares method, transforming the contact migration phenomenon of the mechanical structure into a continuous mathematical index that can be solved by a computer. The specific operation process is as follows:

[0057] The system extracts the dimension size passed in step one as follows: The driving-end two-dimensional vector matrix va is used as a stacked driving-end vector matrix ma in subsequent batch operations. Similarly, the non-driving-end two-dimensional vector matrix vb is extracted as a stacked non-driving-end vector matrix mb. A regularization constant rc and a second-order identity matrix im are also introduced for later use.

[0058] Specifically, under normal operating conditions, the vibration response of the bearing housings at both ends of the spindle satisfies a relatively stable linear transmission relationship. The instantiated two-dimensional vector matrix is ​​used as a stacked matrix in the calculation, aiming to accelerate the solution of the inter-end transmission law through batch matrix operations. This operation effectively avoids redundant reallocation and repeated copying of computer memory. The regularization constant rc is introduced to prevent the covariance matrix from approaching singularity when the amplitude within the sliding window is extremely small or the correlation between the two ends is extremely strong. This underlying configuration greatly avoids numerical divergence or error crashes during subsequent matrix inversion operations, ensuring the high robustness of the industrial online monitoring system code.

[0059] It should be noted that, in order to achieve excellent numerical stability without masking the true inter-end differences, the range of values ​​for the regularization constant rc is limited in practice. to In this embodiment, the preferred embodiment is... The second-order identity matrix *im* has elements of 1 on its main diagonal and all other elements of zero. The standard matrix. The stacked matrix extracted in the aforementioned process is a contiguous two-dimensional floating-point array in the memory of the computing device, providing a rigorous computational object for the subsequent closed-form solution of the equations.

[0060] Using the driver-end vector stacking matrix ma, the non-driver-end vector stacking matrix mb, the regularization constant rc, and the second-order identity matrix im, the system solves for the optimal solution using a closed-form formula with a stability term. The inter-terminal linear mapping matrix wm. The calculation formula is:

[0061]

[0062] The above formula indicates that the inter-end linear mapping matrix wm is equal to the intermediate product matrix obtained by multiplying the non-driving end vector stack matrix mb by the transpose of the driving end vector stack matrix ma. The right multiplication is the inverse of a composite matrix, which is composed of the product of the driving end vector stack matrix ma and its own transpose, plus the product of the regularization constant rc and the second-order identity matrix im.

[0063] Specifically, the inter-end linear mapping matrix wm represents the optimal linear transfer weight from the power input side to the non-power input side within the current sliding window. This closed-form solution process uses the least squares rule with a penalty term to fit the spatial transfer structure at both ends of the main shaft under normal rigid constraints, essentially reducing the complexity of the mechanical force transmission path changes to a purely digital spatial mapping relationship.

[0064] It should be noted that the composite matrix inversion operation with the regularization constant rc ensures that even under extremely low signal-to-noise ratio conditions, the computer processor can still output definite and bounded mapping weights. Explicitly specifying the mathematical action of right multiplication, i.e., post-multiplication, strictly defines the non-commutativity of matrix multiplication, effectively avoiding logical ambiguity at the linear algebra code implementation level.

[0065] After obtaining the inter-end linear mapping matrix wm, the system calculates the minimum mean square residual of the inter-end mapping, i.e., the contact migration index ci. The calculation formula is:

[0066]

[0067] The above formula indicates that the contact migration index ci is equal to the sum of the squares of the difference between the corresponding elements of the non-driving end two-dimensional vector matrix vb and the inter-end linear mapping matrix wm multiplied by the corresponding elements of the driving end two-dimensional vector matrix va under all time indices, divided by the total number of sample points np.

[0068] Specifically, when the semi-wedge-entry locking rope contact occurs, the lateral thrust and additional bending moment introduced by the burrs on the rope groove lip instantly rewrite the load path of the spindle assembly, causing a significant deviation in the inter-end transmission relationship within the sliding window. The inter-end linear mapping matrix wm fitted based on normal operating conditions cannot explain the nonlinear deviation, resulting in a large residual when projecting between the ends. The contact migration index ci serves as a quantitative indicator of the residual; a larger value indicates a more unstable inter-end transmission structure. This computational mechanism effectively filters out the common conventional rotational vibration background shared by both ends, purifying the hidden contact migration phenomenon into a continuous scalar characteristic that can be directly judged by the computer.

[0069] It should be noted that the contact migration index ci, as a continuous floating-point scalar reflecting the current time slice state, will be passed to the next level of the computation process as a dedicated and direct input feature, specifically used to update the historical background queue and calculate the dynamic wedge weights. In computer memory allocation, the inter-end linear mapping matrix wm is explicitly instantiated as... A two-dimensional floating-point matrix array.

[0070] Step 3: When a semi-wedge-entry locking rope contact occurs in the main shaft assembly of a mine hoist, the strong contact event characteristics generated can mask or confuse the actual bearing and main shaft fault signals. To ensure the data purity of the subsequent state discrimination model, Step 3 aims to remove the interference caused by the contact event from the main fault discrimination data stream, forming a stable data component resistant to interference, as follows:

[0071] The system maintains a first-in-first-out (FIFO) queue in the computing device's memory to store the contact migration indices of the most recent sliding windows. The system extracts the current window's contact migration index ci from step two and pushes it into this FIFO queue. The system presets the window length of the background cache queue to be qn. When the total number of elements in the queue exceeds the window length qn, the system automatically pops the oldest historical value. Subsequently, the system calculates the arithmetic mean of all contact migration indices in the current queue, denoted as the background mean contact migration index bm.

[0072] Specifically, the process of establishing a dynamically updated historical background queue and performing mean calculation is based on the fact that operating conditions in industrial settings, such as load magnitude and environmental noise, exhibit slow, time-varying drift. Using a fixed threshold to judge contact events is prone to threshold failure. By capturing the recent basic operating status of the spindle device in real time, this dynamic background extraction mechanism substantially enhances the system's adaptability under varying operating conditions, greatly avoiding misjudgments due to baseline drift caused by long-term wear or gradual load changes.

[0073] It should be noted that, to ensure the background mean reflects both short-term operating conditions and is not contaminated by a single abnormal event over a long period, the window length qn is limited to a range of 20 to 60 in practice, with 30 being preferred in this embodiment. With a window step size of one second, thirty windows correspond precisely to approximately thirty seconds of operating background. The contact migration index background mean bm, as a continuous floating-point scalar representing the current macroscopic operating state, provides an objective benchmark for subsequent weight allocation.

[0074] The continuous wedge weight wg, representing the degree of semi-wedge dominance, is calculated using the contact migration index ci of the current window and the background mean bm. The calculation formula is as follows:

[0075]

[0076] The above formula indicates that the continuous wedge weight wg is equal to the contact migration index ci of the current window divided by a denominator, which is composed of the contact migration index ci, the background mean bm, and the minimum positive constant ep.

[0077] Specifically, the core technology behind the continuous wedge weight calculation lies in the fact that traditional anomaly detection logic heavily relies on hard alarm thresholds, resulting in extremely poor robustness when deployed across devices. This solution employs a soft attention mechanism: when the inter-end transfer structure residual of the current window is significantly higher than the recent background mean, it indicates that a semi-wedge phenomenon is dominating the current vibration response. At this point, the calculated continuous wedge weight wg automatically increases to approach one. Under normal operating conditions, this weight remains at a lower level. This calculation mechanism substantially replaces the step-like hard threshold judgment, transforming discrete event detection into smooth probability-level weighting, ensuring high consistency in the underlying logic operations.

[0078] It should be noted that the minimal positive constant ep is used to prevent underlying division-by-zero crashes that occur when the mechanical system is stationary for a long period and all queue elements are zero. The calculated continuous wedge weight wg is a continuous floating-point number ranging from zero to one.

[0079] Using the inter-terminal linear mapping matrix wm generated in step two and the driving end two-dimensional vector matrix va passed in step one, the system predicts the response of the non-driving end point by point according to the time index ti, generating the prediction non-driving end two-dimensional vector matrix pv. The calculation formula is:

[0080]

[0081] The above formula indicates that the predicted non-driving end 2D vector at a specific time index is equal to the inter-end linear mapping matrix wm multiplied by, i.e., left-multiplied by, the driving end 2D vector at the specific time index. Subsequently, the system performs consistency fusion between the driving end 2D vector matrix va and the predicted non-driving end 2D vector matrix pv to generate the inter-end consistent 2D stable component matrix sv. The calculation formula is:

[0082]

[0083] The above formula indicates that the inter-end consistency two-dimensional stable component at a specific time index is equal to half the sum of the driving end two-dimensional vector and the predicted non-driving end two-dimensional vector at that specific time index. Finally, the system introduces a continuous wedge weight wg to dynamically suppress the inter-end consistency two-dimensional stable component matrix sv, generating the final stable input data matrix sd. The calculation formula is:

[0084]

[0085] The above formula indicates that the stable input data under a specific time index is equal to the difference between one and the continuous wedge weight wg, multiplied by the two-dimensional stable component of inter-end consistency under the specific time index.

[0086] Specifically, the core of achieving consistent prediction between endpoints and stable input data generation lies in directly discarding the sliding window where a half-wedge phenomenon occurs, which would lead to data flow disruption. This process extracts the common vibration components at both ends of the main shaft that conform to normal rigid constraints through inter-end mapping projection, filtering out asymmetric strong impacts caused by unilateral wear. A multiplicative suppression operation including a subtraction of a weighting factor term is introduced, which automatically and proportionally reduces the overall amplitude of the stable components when the half-wedge phenomenon dominates. This operation effectively prevents residual high-frequency, high-energy noise from contact events from infiltrating subsequent networks, substantially removing the pollution of the main fault-determining data stream caused by contact events.

[0087] It should be noted that in computer memory allocation, the prediction non-driver two-dimensional vector matrix pv, the inter-end consistency two-dimensional stable component matrix sv, and the stable input data matrix sd are all explicitly instantiated as having a dimension of [missing information]. A two-dimensional floating-point matrix array. The final generated continuous wedge weight wg and stable input data matrix sd will be passed to step four as dedicated and direct data payloads.

[0088] Step Four: When the mine hoist main shaft assembly experiences partial wedging and rope locking contact, a directional force offset will occur at the end. Simultaneously, a genuine bearing or main shaft failure will manifest as a specific mode of vibration response. To achieve precise separation and location of these two types of risks, Step Four aims to generate, in parallel, evidence of the physical structural offset used to locate rope groove defects and evidence of the failure category pointing to bearing or main shaft defects. The specific process is as follows:

[0089] The system extracts the driver-side horizontally normalized sequence px, the driver-side vertically normalized sequence py, the non-driver-side horizontally normalized sequence qx, and the non-driver-side vertically normalized sequence qy from step one. The system first calculates the L2 energy scale for each channel within the current sliding window, obtaining the driver-side horizontal energy scale ea, the driver-side vertical energy scale eb, the non-driver-side horizontal energy scale ec, and the non-driver-side vertical energy scale ed, respectively. The calculation formula is based on the driver-side horizontal energy scale ea as an example:

[0090]

[0091] The above formula indicates that the horizontal energy scale ea at the driving end is equal to the positive square root of the sum of the squared values ​​of the elements of the horizontally normalized sequence px at all time indices ti. Subsequently, the system calculates the end energy differences in the same direction, obtaining the horizontal offset difference bh and the vertical offset difference bv. The calculation formula is:

[0092] bh = ea - ec;

[0093] bv = eb-ed;

[0094] The above formulas indicate that the horizontal offset difference bh equals the horizontal energy scale ea at the driving end minus the horizontal energy scale ec at the non-driving end, and the vertical offset difference bv equals the vertical energy scale eb at the driving end minus the vertical energy scale ed at the non-driving end. Finally, the system uses the horizontal offset difference bh and the vertical offset difference bv to generate a two-dimensional vector bd for the offset direction.

[0095] Specifically, performing independent channel energy scale calculations and spatial offset direction vector extraction has the effect that the lateral thrust and additional bending moment introduced by the semi-wedge locking rope contact alter the distribution of end support reaction forces, significantly enhancing the overall vibration response in a specific direction at a particular end. By calculating the L2 norm summation of each independent channel, the cumulative energy scale of the channel within the current time slice can be effectively quantified. The sign and amplitude of the horizontal offset difference bh and the vertical offset difference bv intuitively reflect the end and direction experiencing greater lateral thrust. This energy difference calculation mechanism substantially transforms the hidden nonlinear contact friction phenomenon into solid structural offset evidence with a clear geometric orientation, providing reliable spatial clues for subsequent positioning of rope groove lip wear or burrs.

[0096] It should be noted that in computer memory allocation, both the horizontal offset difference bh and the vertical offset difference bv are floating-point scalars, and the two-dimensional vector bd of the offset direction is explicitly instantiated as a two-dimensional spatial vector containing two floating-point elements. The two-dimensional vector bd of the offset direction will be passed to step five as exclusive and direct evidence of location.

[0097] The system extracts the dimension size passed in step three as follows: The system inputs a stable input data matrix sd. The system then feeds the entire stable input data matrix sd into a pre-trained fault detection model for feedforward inference. The fault detection model outputs classification results for the preset total number of fault categories fc for the spindle assembly. After normalization and exponential function mapping, the system generates the corresponding probabilities for each fault, forming a fault category probability distribution array pr.

[0098] Specifically, the technical advantage of achieving stable input data inference and fault probability distribution generation lies in the fact that directly inputting contaminated raw vibration data into the classification model can easily lead to the model misclassifying abnormal rope groove contact as internal bearing damage. The aforementioned stable input data matrix sd has already substantially removed asymmetric strong impact interference in the previous stage of processing. Feeding the stable input data matrix sd, after filtering out contact interference, into the fault detection model based on the fusion of physical information and data-driven approaches allows the classifier to highly focus on robust features caused by real mechanistic defects such as bearing inner ring spalling or spindle loosening. This dual-line decoupled feedforward inference process greatly avoids cross-classification misclassification and ensures a high degree of confidence in the fault category probability distribution.

[0099] It should be noted that, in specific implementation, the fault detection model is preferably instantiated as a lightweight one-dimensional convolutional neural network. This network sequentially includes: a first one-dimensional convolutional layer that maps two-dimensional input channels to sixteen dimensions, with a kernel length preferably of nine; a second one-dimensional convolutional layer that outputs thirty-two-dimensional features and cascades max pooling operations; a third one-dimensional convolutional layer that outputs sixty-four-dimensional features and cascades global average pooling operations; and a fully connected layer that finally maps to the total number of fault categories, fc. The preset total number of fault categories, fc, is determined by the actual training data annotation system in the industrial field. In this embodiment, it preferably covers normal conditions, bearing faults, spindle misalignment / loosening faults, and drum eccentricity / imbalance faults. In computer memory allocation, the fault category probability distribution array pr is explicitly instantiated as a one-dimensional floating-point array with a length equal to the total number of fault categories, fc, where all elements in the array have values ​​greater than or equal to zero and their sum is strictly equal to one. The generated fault category probability distribution array pr serves as direct evidence characterizing the confidence level of bearing and spindle defects. Together with the two-dimensional vector bd of the bias direction, it is incorporated into the next step, step five, for dual priority competition and dynamic instruction allocation.

[0100] In step five, the main shaft assembly of the mine hoist exhibits a unique vibration evolution pattern when the load path is rewritten, and bearing or main shaft failures also manifest as specific energy distribution patterns. To complete the monitoring loop at the terminal, and dynamically determine and output the priority of maintenance instructions based on the current data stream status, step five aims to perform bidirectional priority contention calculation and dynamic distribution of structured maintenance instructions. The specific process is as follows:

[0101] The system extracts the continuous wedge weight wg passed in step three, and simultaneously extracts the horizontal bias difference bh, vertical bias difference bv, and fault category probability distribution array pr passed in step four. The system first calculates the bias strength bt using the horizontal bias difference bh and the vertical bias difference bv. The calculation formula is:

[0102]

[0103] The above formula indicates that the offset strength bt is equal to the positive square root of the sum of the squares of the horizontal offset difference bh and the squares of the vertical offset difference bv. Subsequently, the system calculates the rope groove routing inspection priority pg and the bearing spindle maintenance priority pb, respectively. The calculation formulas are as follows:

[0104]

[0105] The above formula indicates that the priority of rope groove rope arrangement inspection pg is equal to the continuous wedging weight wg multiplied by the bias strength bt.

[0106]

[0107] The above formula indicates that the bearing spindle maintenance priority pb is equal to the difference between the continuous wedging weight wg and the maximum element value in the fault category probability distribution array pr.

[0108] Specifically, the bias strength calculation and dual-priority quantization are performed because the end-bias risk caused by load path rewriting depends not only on the degree of matching of feature patterns (i.e., the weight magnitude) but also on the severity of the actual bias response. Multiplying the continuous wedge weights by the bias strength allows for a reasonable quantification of the abnormal mechanical force risk caused by frictional contact. Simultaneously, introducing a factor of 1 minus the continuous wedge weights suppresses the maximum failure probability, effectively mitigating the overconfidence tendency that the classifier might exhibit when contact events dominate. This dual-line dynamic competition mechanism substantially replaces the traditional single-dimensional hard threshold alarm, significantly improving the reliability of maintenance decisions under complex operating conditions.

[0109] It should be noted that in computer memory allocation, the bias strength bt, the rope groove rope inspection priority pg, and the bearing spindle maintenance priority pb are all explicitly instantiated as scalar floating-point values, providing clear numerical judgment basis for subsequent logic control units.

[0110] Based on the calculated dual priorities, the system makes a comparative logical decision. If the rope groove alignment inspection priority pg is greater than or equal to the bearing spindle maintenance priority pb, the system determines that the risk of rope groove contact abnormality is higher and assigns the maintenance instruction type it as a rope groove inspection instruction. Conversely, if the rope groove alignment inspection priority pg is less than the bearing spindle maintenance priority pb, the system determines that the risk of mechanism damage is higher and assigns the maintenance instruction type it as a maintenance instruction. Simultaneously, the system extracts the fault name corresponding to the index with the maximum value in the fault category probability distribution array pr and assigns it as the suggested fault category ft. Finally, the system packages the aforementioned data and reports it through a preset communication channel.

[0111] Specifically, the core technology for achieving dynamic instruction generation and structured result reporting lies in identifying that when contact migration is the dominant phenomenon, the system prioritizes outputting an action list including the inspection and repair of the drum rope groove lip and transition zone, and provides specific end and direction positioning clues based on the offset difference. When a mechanistic fault is the dominant phenomenon, the system provides specific bearing or spindle maintenance instructions. This step completes the decision-making closed loop between the data-driven underlying algorithm and actual physical operation and maintenance, substantially reducing the ineffective downtime costs caused by incorrect maintenance due to monitoring system misjudgments.

[0112] It should be noted that, to ensure seamless information flow in the industrial field, the generated structured result package preferably includes data fields such as the timestamp of the current window's start time, continuous wedge weight wg, horizontal offset difference bh, vertical offset difference bv, rope groove rope arrangement inspection priority pg, bearing spindle maintenance priority pb, suggested fault category ft, and maintenance instruction type it. After the data is packaged, the system will report it to the host computer or maintenance cloud platform through standard industrial IoT protocol channels such as object linking and embedding unified architecture, message queue telemetry transmission, or presentation layer state transition application programming interfaces.

[0113] Furthermore, a data-driven fault detection system for the main shaft assembly of a mine hoist is proposed to implement any of the above-mentioned data-driven fault detection methods for the main shaft assembly of a mine hoist, including:

[0114] The data acquisition module is used to extract the decentralized sequence of mechanical vibration signals, calculate the median absolute deviation of the decentralized sequence to generate a robust scaling factor, and use the robust scaling factor to scale and combine the output two-dimensional vector matrix of the driving end and the two-dimensional vector matrix of the non-driving end.

[0115] The wedge characterization module is used to solve the inter-end linear mapping matrix based on the two-dimensional vector matrix of the driving end and the two-dimensional vector matrix of the non-driving end, and to calculate the minimum mean square residual to generate the contact migration index.

[0116] The load decoupling module is used to calculate the arithmetic mean of the contact migration index and combine it with the contact migration index to calculate the continuous wedge weight. Based on the inter-end linear mapping matrix and the two-dimensional vector matrix of the driving end, it predicts and generates an inter-end consistent two-dimensional stable component matrix. After continuous wedge weight multiplication suppression, a stable input data matrix is ​​generated.

[0117] The state discrimination module is used to calculate the end energy difference based on the two-dimensional vector matrix of the driving end and the two-dimensional vector matrix of the non-driving end, and input the stable input data matrix into a one-dimensional convolutional neural network to generate a fault category probability distribution array.

[0118] The instruction generation module is used to calculate the bias strength using the end energy difference, and to calculate the rope groove rope arrangement inspection priority and bearing spindle maintenance priority by combining the continuous wedge weight, bias strength and fault category probability distribution array, and output the operation and maintenance instruction type.

[0119] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A data-driven fault detection method for the main shaft assembly of a mine hoist, characterized in that, include: The decentralized sequence of the mechanical vibration signal is extracted, the median absolute deviation of the decentralized sequence is calculated to generate a robust scaling factor, and the robust scaling factor is used to scale and combine the output two-dimensional vector matrix of the driving end and the two-dimensional vector matrix of the non-driving end. Based on the two-dimensional vector matrix of the driving end and the two-dimensional vector matrix of the non-driving end, the linear mapping matrix between the ends is solved, and the minimum mean square residual is calculated to generate the contact migration index. The arithmetic mean of the contact migration index is calculated and the continuous wedge weight is calculated in combination with the contact migration index. Based on the inter-end linear mapping matrix and the driving end two-dimensional vector matrix, the inter-end consistency two-dimensional stable component matrix is ​​predicted and generated. The stable input data matrix is ​​generated by the continuous wedge weight multiplication suppression. The end energy difference is calculated based on the two-dimensional vector matrix of the driving end and the two-dimensional vector matrix of the non-driving end. The stable input data matrix is ​​then input into a one-dimensional convolutional neural network to generate a fault category probability distribution array. The bias strength is calculated using the end energy difference. The priority of rope groove rope arrangement inspection and bearing spindle maintenance is calculated by combining the continuous wedge weight, bias strength and fault category probability distribution array, and the operation and maintenance instruction type is output.

2. The data-driven fault detection method for the main shaft device of a mine hoist as described in claim 1, characterized in that, The specific process for extracting the decentralized sequence of mechanical vibration signals includes: Simultaneously capture the original four-channel acceleration sequence within a sliding window; The original four-channel acceleration sequence is subjected to invalid value replacement and saturation overflow value truncation to generate a repaired original four-channel acceleration sequence. If there are invalid values ​​in the original four-channel acceleration sequence, the invalid values ​​are replaced by the median of the valid values ​​in the corresponding channel of the sliding window. If the original four-channel acceleration sequence has saturation overflow values, the saturation overflow values ​​are truncated to the upper limit of the full scale. Calculate the window size of the repaired four-channel original acceleration sequence, and subtract the window size from each element of the repaired four-channel original acceleration sequence to generate a decentralized sequence.

3. The data-driven fault detection method for the main shaft device of a mine hoist as described in claim 1, characterized in that, The specific process of calculating the median absolute deviation of the decentralized sequence to generate a robust scaling factor, and then scaling and combining the output of the driving-end two-dimensional vector matrix and the non-driving-end two-dimensional vector matrix using the robust scaling factor includes: Calculate the median of the set of absolute values ​​of each element of the decentralized sequence of each channel minus the median of the overall decentralized sequence, and generate the median absolute deviation of each channel. The arithmetic mean of the absolute deviations of the median in the horizontal direction and the absolute deviations of the median in the vertical direction at the same end is used to generate the mean of the component deviations. The robust scaling factor is generated by summing the mean of the component deviations with the smallest positive constant and then taking the square root. The decentralized sequence is divided by the robust scaling factor of its respective end to generate a normalized sequence. The normalized sequences under the same time index of the driving end are spatially combined to generate a two-dimensional vector matrix of the driving end. The normalized sequences under the same time index of the non-driving end are spatially combined to generate a two-dimensional vector matrix of the non-driving end.

4. The data-driven fault detection method for the main shaft device of a mine hoist as described in claim 1, characterized in that, The specific process of solving the inter-end linear mapping matrix based on the two-dimensional vector matrix of the driving end and the two-dimensional vector matrix of the non-driving end includes: The two-dimensional vector matrix of the driving end and the two-dimensional vector matrix of the non-driving end are respectively treated as the driving end vector stacking matrix and the non-driving end vector stacking matrix. Multiply the non-driving-end vector stack matrix by the transpose of the driving-end vector stack matrix to generate an intermediate product matrix. The composite matrix is ​​generated by multiplying the stacked matrix of the driving vectors by the transpose of the stacked matrix of the driving vectors, and then adding the product of the regularization constant and the second-order identity matrix. The inverse of the composite matrix is ​​obtained by multiplying the intermediate product matrices by the end product matrix, thus generating the inter-terminal linear mapping matrix.

5. The data-driven fault detection method for the main shaft device of a mine hoist as described in claim 1, characterized in that, The specific process of calculating the minimum mean square residual to generate the contact migration index includes: Calculate the sum of the squares of the 2-norm of the differences between the products of the corresponding elements of the non-driving end 2D vector matrix and the inter-end linear mapping matrix multiplied by the corresponding elements of the driving end 2D vector matrix under all time indices. The contact migration index is generated by dividing the sum of the squares of the L2 norm by the total number of sample points.

6. The data-driven fault detection method for the main shaft device of a mine hoist as described in claim 1, characterized in that, The specific process of calculating the arithmetic mean of the contact migration index and combining it with the contact migration index to calculate the continuous wedge weight includes: Maintain the first-in-first-out queue and push the contact migration index of the current window into the first-in-first-out queue; When the total number of elements in the first-in-first-out queue exceeds the window length, the oldest contact migration index in the first-in-first-out queue is popped. Calculate the arithmetic mean of all contact migration indices within the first-in, first-out queue; The continuous wedge weight is generated by dividing the current window's contact migration index by the sum of the current window's contact migration index, the arithmetic mean of the contact migration indices, and the smallest positive constant.

7. The data-driven fault detection method for the main shaft device of a mine hoist as described in claim 1, characterized in that, The specific process of generating a stable two-dimensional component matrix with inter-end consistency based on the inter-end linear mapping matrix and the driving end two-dimensional vector matrix, and then generating a stable input data matrix through continuous wedge-in weight multiplication suppression, includes: Multiply the inter-end linear mapping matrix by the driving end two-dimensional vector matrix to generate the prediction non-driving end two-dimensional vector matrix. The inter-end consistency two-dimensional stable component matrix is ​​generated by calculating half the sum of the two-dimensional vector matrix of the driving end and the predicted two-dimensional vector matrix of the non-driving end. Calculate the difference between the numerical value 1 and the continuous wedge weights, and multiply the difference by the inter-end consistency two-dimensional stable component matrix to generate a stable input data matrix.

8. The data-driven fault detection method for the main shaft device of a mine hoist according to claim 1, characterized in that, The specific process of calculating the end-effector energy difference based on the two-dimensional vector matrix of the driving end and the two-dimensional vector matrix of the non-driving end, and then inputting the stable input data matrix into a one-dimensional convolutional neural network to generate a fault category probability distribution array includes: Calculate the positive square root of the sum of the squared values ​​of the elements of each channel under all time indices in the two-dimensional vector matrix of the driving end and the two-dimensional vector matrix of the non-driving end, and generate the horizontal energy scale of the driving end, the vertical energy scale of the driving end, the horizontal energy scale of the non-driving end, and the vertical energy scale of the non-driving end, respectively. The horizontal offset difference is generated by subtracting the horizontal energy scale of the non-drive end from the horizontal energy scale of the drive end, and the vertical offset difference is generated by subtracting the vertical energy scale of the non-drive end from the vertical energy scale of the drive end. The horizontal offset difference and the vertical offset difference together constitute the end energy difference. The one-dimensional convolutional neural network sequentially includes a first one-dimensional convolutional layer that maps two-dimensional input channels to sixteen dimensions and has a convolutional kernel length of nine, a second one-dimensional convolutional layer that outputs thirty-two-dimensional features and cascades max pooling operations, a third one-dimensional convolutional layer that outputs sixty-four-dimensional features and cascades global average pooling operations, and a fully connected layer that maps to the total number of fault categories. A stable input data matrix is ​​fed into a one-dimensional convolutional neural network for feedforward inference to output classification results. The classification results are then mapped using a normalized exponential function to generate a fault category probability distribution array.

9. The data-driven fault detection method for the main shaft device of a mine hoist according to claim 8, characterized in that, The specific process of calculating the offset strength using the end energy difference, and combining the continuous wedging weight, offset strength, and fault category probability distribution array to calculate the rope groove rope arrangement inspection priority and bearing spindle maintenance priority, and outputting the maintenance instruction type includes: The positive square root of the sum of the squares of the horizontal and vertical offset differences is used to generate the offset strength. The continuous wedging weight is multiplied by the bias strength to generate the rope groove rope arrangement inspection priority; Calculate the difference between the value 1 and the continuous wedge weight, and multiply the difference between the value 1 and the continuous wedge weight by the maximum element value in the fault category probability distribution array to generate the bearing spindle maintenance priority. If the priority of rope groove rope arrangement inspection is greater than or equal to the priority of bearing spindle maintenance, the maintenance instruction type is determined to be rope groove inspection instruction. If the priority of rope groove rope arrangement inspection is lower than the priority of bearing spindle maintenance, the maintenance instruction type is determined to be a maintenance instruction.

10. A data-driven fault detection system for the main shaft assembly of a mine hoist, used to implement the data-driven fault detection method for the main shaft assembly of a mine hoist as described in any one of claims 1-9, characterized in that, include: The data acquisition module is used to extract the decentralized sequence of mechanical vibration signals, calculate the median absolute deviation of the decentralized sequence to generate a robust scaling factor, and use the robust scaling factor to scale and combine the output two-dimensional vector matrix of the driving end and the two-dimensional vector matrix of the non-driving end. The wedge characterization module is used to solve the inter-end linear mapping matrix based on the two-dimensional vector matrix of the driving end and the two-dimensional vector matrix of the non-driving end, and to calculate the minimum mean square residual to generate the contact migration index. The load decoupling module is used to calculate the arithmetic mean of the contact migration index and combine it with the contact migration index to calculate the continuous wedge weight. Based on the inter-end linear mapping matrix and the two-dimensional vector matrix of the driving end, it predicts and generates an inter-end consistent two-dimensional stable component matrix. After continuous wedge weight multiplication suppression, a stable input data matrix is ​​generated. The state discrimination module is used to calculate the end energy difference based on the two-dimensional vector matrix of the driving end and the two-dimensional vector matrix of the non-driving end, and input the stable input data matrix into a one-dimensional convolutional neural network to generate a fault category probability distribution array. The instruction generation module is used to calculate the bias strength using the end energy difference, and to calculate the rope groove rope arrangement inspection priority and bearing spindle maintenance priority by combining the continuous wedge weight, bias strength and fault category probability distribution array, and output the operation and maintenance instruction type.

Images