Medical device health state predictive maintenance inspection method based on digital twinning

Abstract

The application discloses a kind of based on digital twinning medical equipment health state predictive maintenance inspection method, comprising: according to patient examination task scheduling plan and equipment historical load spectrum generation equipment operating condition sequence;According to the equipment operating condition sequence and the fatigue cumulative thermal stress peak value predicted by twin simulation, the equipment health state evolution model is constructed;According to the equipment health state evolution model, the maintenance demand priority of non-shutdown detection item in inspection work order is identified;Based on the maintenance demand priority and the time window constraint, the non-shutdown detection item is dynamically scheduled to generate an optimized maintenance scheme.The application deeply correlates the actual load history of the equipment with the deterioration trend through digital twinning technology, and accurately embeds the key state detection task into the highly fragmented equipment idle window without affecting normal diagnosis and treatment, solving the problem of maintenance lag caused by the inability to insert an effective detection window during the peak period of high utilization rate imaging equipment.

Description

Technical Field

[0001] This application relates to the field of equipment operation and maintenance technology, specifically to a predictive maintenance and inspection method for the health status of medical equipment based on digital twins. Background Technology

[0002] Given the current high-load operation of medical equipment, imaging equipment maintenance strategies face significant challenges. Traditional maintenance and inspection methods typically employ fixed cycles or reactive mechanisms, which struggle to adapt to constantly changing patient examination schedules and equipment operating statuses. Especially during peak periods, high equipment utilization and a lack of effective inspection windows often lead to the delay or neglect of non-downtime inspection items (such as radiofrequency coil performance monitoring and cooling system pressure trend analysis), resulting in maintenance lags that impact equipment reliability and treatment quality. To address this issue, there is an urgent need for a method that can dynamically optimize the scheduling of inspection tasks by combining equipment operating status, task scheduling plans, and simulation data, thereby improving the timeliness and effectiveness of medical equipment maintenance. Summary of the Invention

[0003] In view of this, the present disclosure provides a predictive maintenance and inspection method for the health status of medical devices based on digital twins, which at least partially solves the problems existing in the prior art.

[0004] A predictive maintenance and inspection method for the health status of medical devices based on digital twins includes: Generate equipment operating condition sequence based on patient examination task scheduling plan and equipment historical load spectrum; Based on the equipment operating condition sequence and the fatigue cumulative thermal stress peak predicted by twin simulation, a health status evolution model of the equipment is constructed. Based on the equipment health status evolution model, identify the maintenance requirement priority of non-downtime inspection items in the inspection work order; Based on the maintenance requirement priority and timing window constraints, non-downtime detection items are dynamically scheduled to generate an optimized maintenance plan.

[0005] In one specific implementation, generating the equipment operating condition sequence based on the patient examination task scheduling plan and the equipment's historical load spectrum includes: Obtain the daily usage time and task category for each medical device in the patient examination task scheduling plan; Extract the kV / mAs sequence distribution and time axis data of scanning bed cycle frequency from the historical load spectrum of the equipment; Synchronize and align task time periods with load spectrum data based on timestamps and construct multi-dimensional time series feature vectors; The operating intensity index I is determined by the following formula: I = α·(1 / ΔT+β·R)+γ·F, where ΔT represents the task interval time, R represents the scanning bed cycle frequency, F is the peak frequency in kV / mAs, and α, β, and γ are weighting coefficients to quantify the current load status of the equipment.

[0006] In one specific implementation, the step of constructing a health state evolution model of the equipment based on the equipment operating condition sequence and the fatigue cumulative thermal stress peak predicted by twin simulation includes: Extract the load characteristic parameters for each time period in the operating condition sequence, including the maximum thermal stress σ_max, the temperature rise rate dT / dt, and the number of fatigue cycles N; The cumulative thermal stress σ_cum in each time period is obtained by calling the finite element simulation model in the digital twin. The equipment health index H is calculated using the following weighted formula: H=Σ(σ_cum / σ_yield)^p×w_t, where σ_yield is the material yield strength, p is the nonlinear decay exponent, and w_t is the time weight; The H value is compared with the equipment safety threshold to determine whether the equipment's health status is in a critical region.

[0007] In one specific implementation, the step of identifying the maintenance requirement priority of non-downtime inspection items in the inspection work order based on the equipment health status evolution model includes: Obtain the equipment health index H and divide it into three priority levels according to the interval; Define the sensitivity factor S_i for the detection item, including the rate of change of Q value during online monitoring and the slope of the helium pressure trend; The priority index P_i for each non-stop detection task is calculated using the formula P_i=(S_i×H) / K, where K is an empirical adjustment factor. Sort by P_i, output a priority list and mark the high-priority detection tasks.

[0008] In one specific implementation, the dynamic scheduling of non-downtime detection items based on the maintenance requirement priority and timing window constraints includes: Define the tolerable delay time T_tol for inspection work orders during peak business periods; Obtain available free windows from the device and divide them into multiple time-series candidate windows by time period; The adaptation score for each window is calculated using the formula W_score=(T_free×S_prio) / (T_delay), where T_free is the window duration, S_prio is the task priority score, and T_delay is the actual allowed delay time. The deployment schedule is arranged by selecting the most suitable window based on maximizing W_score.

[0009] In one specific implementation, defining the tolerable delay time T_tol for the detection work order during peak business periods includes: Based on the hospital's average daily patient volume C for imaging equipment and the service response standard SLA, the range of peak business periods is defined. Set T_tol=T_service_window×μ / (1+e^(k×(C-C_threshold))), where T_service_window is the average available idle time, μ is the delay coefficient, k is the turning point factor, and C_threshold is the critical number of patients. If C ≤ C_threshold, T_tol is set to the normal range; otherwise, T_tol is gradually reduced as the number of patients increases to reduce the risk of delays.

[0010] In one specific implementation, the acquisition device can use an idle window to form multiple time-series candidate windows by dividing them into time periods, including: Record the device's daily operation log and extract all time segments during periods of no task. After standardizing and normalizing each blank time segment, candidate windows that meet the minimum length L_min are selected. Define the window matching index M=min(Δt / τ,1), where Δt is the expected time for target detection and τ is the window coverage time to ensure effective scheduling feasibility; The remaining candidate windows are ranked and scored according to their time constraints and scheduling flexibility.

[0011] In one specific implementation, the adjustment of task execution order based on priority indicators further includes: Establish a dependency graph G among non-shutdown detection items; The weight vector W_p is constructed using the AHP (Analytic Hierarchy Process) to consider the complexity and risk impact of the detection project. Combining the H value and task priority P_i, the task completion probability P_total = α·P_i + β·H + γ·D is defined, where D is the dependency condition satisfaction rate, and α, β, and γ are decision weights. Dynamic programming algorithms are used to select the globally optimal task execution order to improve scheduling rationality.

[0012] In one specific implementation, the step of selecting the globally optimal task execution order using a dynamic programming algorithm includes: Establish the state transition matrix F, and establish the state cost function Cost(i,j) based on time conflicts and resource requirements; The optimization objective is defined as maximizing the completion efficiency, i.e., MaxΣ(P_i) t_i), where t_i is the duration required for the task; A genetic algorithm is introduced, with the population size P_size and the number of evolutions Gen_count set; Each iteration generates a chromosome sequence with high fitness until the convergence threshold is met or the maximum number of iterations is reached.

[0013] This invention, by deeply integrating patient examination task scheduling plans, historical equipment load spectra, and the simulation and prediction capabilities of digital twins for fatigue-accumulated thermal stress, constructs a dynamic evolution model that accurately reflects the trend of equipment health degradation. This transforms the traditional passive, fixed-cycle inspection mode into a predictive maintenance paradigm based on actual load history and real-time risk quantification. Furthermore, by establishing a quantitative coupling relationship between equipment health indices and sensitive factors of inspection items, it achieves precise prioritization of the urgency of non-downtime inspection tasks such as online monitoring of RF coil Q values ​​and cold head helium pressure trend analysis. Moreover, by combining the tolerable delay time during peak business periods with the time constraints of fragmented idle windows, key inspection items are precisely embedded into effective time gaps through dynamic programming or greedy scheduling algorithms. This solution fundamentally addresses the industry pain point of long-term lag in status monitoring of high-utilization medical imaging equipment during peak diagnostic periods due to the inability to find suitable detection windows. It enables early detection and intervention of sub-health conditions of equipment without affecting normal patient examination procedures, significantly reducing the probability of sudden downtime, ensuring the continuity of medical services, and maximizing equipment availability and full lifecycle maintenance benefits. It has outstanding clinical practical value and promising prospects for engineering promotion. Attached Figure Description

[0014] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the various drawings denote the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings depict only some embodiments disclosed in this application and should not be construed as limiting the scope of this application.

[0015] Figure 1 This is a flowchart of a predictive maintenance and inspection method for the health status of medical equipment based on digital twins. Detailed Implementation

[0016] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0017] Next, referring to the accompanying drawings, the present invention provides a predictive maintenance and inspection method for the health status of medical devices based on digital twins. Figure 1 As shown, the method of the present invention includes: S101: Generate a sequence of equipment operating conditions based on the patient examination task scheduling plan and the equipment's historical load spectrum.

[0018] Specifically, future patient examination scheduling data is obtained from the hospital's information system and combined with the equipment's original load records (e.g., kV / mAs sequences, scan bed cycle frequency) for data fusion analysis. The time axis is divided into a series of time intervals, and the equipment's operating status is statistically analyzed within each time interval, such as high-load periods or frequent start-up periods. For example, in the operation of CT equipment in a tertiary hospital, by integrating the daily examination appointments for each machine, its peak usage range between 8:00 AM and 10:00 PM is determined; simultaneously, the kV / mAs change trend during each scan is filtered from the load records of the past three months, and combined with the scan bed displacement frequency to form a comprehensive load spectrum. The data obtained in this way can be used as input for the next part of the health management model.

[0019] Taking a 64-slice spiral CT scanner as an example, the daily patient examination schedule is extracted from the Radiology Information Management System (RIS), which includes detailed task time periods and examination types such as "8:00-8:10 chest plain scan" and "8:10-8:25 abdominal enhanced scan". At the same time, the historical load spectrum database of the equipment is called up, and the time axis data of the kV and mAs sequence distribution statistical characteristics and the scanning bed entry and exit frequency are extracted for each type of examination. For example, the typical parameters for chest plain scan are tube voltage of 120kV, average tube current of 200mAs, and scanning bed cycle frequency of about 0.5 times / second. Next, with a time step of 1 second, the planned task period is timestamped and aligned with the historical load spectrum parameters: For any second, if the current time is within the planned period of a certain inspection task, the task interval ΔT is set to zero, the scanning bed cycle frequency R is taken as the statistical average value corresponding to that inspection type, and the frequency of pulses whose mAs value exceeds 1.5 times the average mAs value within that second is counted as the peak frequency F; if the current time is in an idle state, ΔT is the cumulative duration of the idle segment from the beginning to the current time, and R and F are both zero. Based on this, the equipment load intensity is calculated second by second using the operating intensity index formula I=α·(1 / ΔT+β·R)+γ·F, where the weighting coefficients α, β, and γ are calibrated to 0.4, 0.3, and 0.3, respectively. Finally, the timestamp of each time step, the operating intensity index I, the scanning bed cycle frequency R, and the peak frequency F are assembled into a multi-dimensional feature vector in chronological order. The continuous sequence of all vectors constitutes the equipment operating condition sequence of the CT equipment in a future scheduling cycle.

[0020] S102: Based on the equipment's operating condition sequence and the fatigue cumulative thermal stress peak predicted by twin simulation, a health status evolution model for the equipment is constructed. This model uses time series as the dimension to build a multi-dimensional state assessment system. For example, on a specific medical imaging device, the heat distribution and material fatigue curve under high load conditions are simulated, and then combined with actual measurement results such as vibration data and current fluctuations of real equipment to form a comparative analysis of simulation results. Specifically, CFD (Computational Fluid Dynamics) is used to simulate the temperature difference distribution of internal components of the equipment, and the maximum allowable stress range of each part is determined by combining knowledge of materials science. The thermal stress value is updated in real time during the simulation process. When the cumulative stress approaches the threshold, an early warning signal is triggered, and the state change trend within that time period is recorded to help determine the urgency and necessity of subsequent maintenance.

[0021] Taking this 64-slice CT scanner as an example, for its core vulnerable component, the X-ray tube anode target plate, load characteristic parameters are extracted from the previously generated equipment operating condition sequence in 15-minute intervals. These include the maximum thermal stress σ_max on the target plate surface obtained through thermodynamic calculations, the temperature rise rate dT / dt estimated from the temperature difference between adjacent time intervals, and the number of fatigue cycles N statistically derived from the temperature cycle of the target plate using the rainflow counting method. The existing digital twin of the CT scanner is then invoked. This twin contains a deep neural network-based reduced-order surrogate model based on the finite element model of the anode target plate. After inputting the aforementioned characteristic parameters, the cumulative thermal stress σ_cum generated by thermomechanical fatigue of the target plate within that time interval can be directly output. Furthermore, the weighted health index formula H=Σ( The evolution trend of the health status of the equipment is quantified by σ_cum / σ_yield)^p×w_t, where σ_yield is taken as the yield strength of the tungsten-rhenium alloy material of the anode target plate, which is 450MPa. The nonlinear decay index p is set to 2 to reflect the nonlinear acceleration effect of damage under high stress level. The time weight w_t is distributed according to exponential decay, that is, the weight of the current 15-minute period is 1.0, and the weight of each subsequent period is multiplied by the decay factor of 0.95, thus accumulating the damage contribution of each period in the past 24 hours. Finally, a continuous curve of the health index H changing with time is obtained. When the H value exceeds the preset safety threshold of 0.75, the equipment is determined to enter a critical state that requires priority maintenance. This dynamic curve constitutes the health status evolution model of the CT equipment.

[0022] S103: Based on the equipment health status evolution model, identify the maintenance requirement priority of non-downtime inspection items in the inspection work order. Here, several key inspection indicators need to be defined, such as online monitoring of the RF coil Q value and cold head helium pressure trend analysis. By modeling and analyzing the weights of different inspection items in the equipment aging model, it can be clarified which indicator changes have a greater impact on equipment stability. For example, assuming that the RF coil temperature rises due to high-frequency scanning within a certain period, the RF coil Q value may decrease accordingly; simultaneously, the cold head helium pressure also becomes abnormal under high-load operating conditions. By comparing the predictive impact coefficients of the two, it is found that the RF coil has a higher degree of degradation, therefore it will be given a higher weight in the priority ranking, thus ensuring that it is executed first in the maintenance plan.

[0023] Based on the aforementioned CT equipment health status evolution model, the current output health index H is 0.72 (close to the safe threshold of 0.75). Three non-stop inspection work orders for the CT equipment are retrieved from the computerized maintenance management system: online monitoring of tube voltage waveform, cooling oil flow trend analysis, and collimator photosensitive array sensitivity drift detection. For each inspection, the sensitivity factor S_i is calculated: S_kV for online monitoring of tube voltage waveform is the average percentage increase of the daily voltage ripple peak value deviating from the baseline value over the past week, calculated to be 0.8; S_oil for cooling oil flow trend analysis is the absolute value of the slope obtained by least-squares linear fitting of the daily average oil flow rate, which reaches 0.6 due to recent slight wear of the oil pump; S_col for collimator sensitivity detection has a stable historical data with a change rate of only 0.2. Substituting the above parameters into the priority calculation formula P_i=(S_i×H) / K, where the normalization constant K is set to 0.1, the priority indices for pipe voltage waveform monitoring are calculated as follows: P_kV=(0.8×0.72) / 0.1=5.76, cooling oil flow analysis P_oil=(0.6×0.72) / 0.1=4.32, and collimator detection P_col=(0.2×0.72) / 0.1=1.44. Priority levels are determined based on P_i values: P_kV≥5.0 is high priority, P_oil between 3.0 and 5.0 is medium priority, and P_col below 3.0 is low priority. Finally, a priority list of detection tasks is generated, ordered in descending order of P_i, with pipe voltage waveform online monitoring clearly marked as a high-priority maintenance item requiring scheduling in the next available window.

[0024] S104: Based on the maintenance requirement priority and timing window constraints, the non-downtime detection items are dynamically scheduled to generate an optimized maintenance plan.

[0025] Specifically, for example, in a radiology department, if multiple patients have CT scan appointments scheduled for the day, and one of the machines is operating at low intensity, the system can detect that it has sufficient idle time at that moment to perform certain maintenance checks, and therefore schedule an online monitoring task for the radiofrequency coil Q value. When the machine is under high load, the system automatically adjusts the timing of this non-stop monitoring task to select a relatively idle period for the machine. This minimizes the impact on normal diagnostic and treatment processes, improving the continuity and efficiency of equipment operation and maintenance.

[0026] Taking this CT scanner as an example, idle time slots are extracted from the daily patient examination schedule, resulting in three candidate idle windows that meet the minimum examination duration of 5 minutes: 9:15-9:25 AM (10 minutes), 11:30-11:40 AM (10 minutes), and 2:00-2:15 PM (15 minutes). The current patient volume C is 60 cases, exceeding the critical value of 50. Based on the tolerable delay time formula, T_tol is calculated to be approximately 4.8 minutes. Therefore, T_delay = 4.8 minutes is adopted for all tasks. Starting with the high-priority task, the online monitoring of the tube voltage waveform is expected to take 8 minutes. The adaptation scores for each window are calculated: 9:15 window W_score = (10 × 5.76) / 4.8 = 12.0, 11:30 window also has a score of 12.0, and 14:00 window W_score = (15 × 5.76) / 4.8 = 18.0. Therefore, it is prioritized for deployment in the highest-scoring 14:00-14:15 window, actually occupying 14:00-14:08. Next, the medium-priority cooling oil flow analysis is processed (taking 7 minutes). The remaining windows are reassessed. After deducting possible preparation time, the 9:15 window still has approximately 9 minutes remaining, W_score = (9 × 4.32) / 4.8 = 8.1. The 11:30 window is the same. The earlier 9:15-9:24 window is selected for execution. At this time, the low-priority collimator sensitivity test (taking 6 minutes) was only available at the 11:30 window, with a score of W_score = (10 × 1.44) / 4.8 = 3.0. Although the score was low, it was still acceptable, so it was scheduled to be performed between 11:30 and 11:36. This led to an optimized maintenance plan: tube voltage waveform monitoring at 14:00, cooling oil flow analysis at 9:15, and collimator testing at 11:30. All tests were precisely embedded during the fragmented idle periods of high-utilization equipment, completely avoiding conflicts with patient examinations.

[0027] This invention deeply integrates patient examination task scheduling plans with historical equipment load spectra to generate a refined operating condition sequence. It then drives digital twin simulation to accurately predict fatigue-accumulated thermal stress, thereby constructing a health status evolution model that realistically reflects equipment degradation trends. This overcomes the lag and blindness of traditional fixed-cycle or experience-based maintenance judgments. Furthermore, by using dual quantification methods of health indices and sensitive factors, the priority of maintenance needs for non-downtime inspection items is dynamically identified. Combined with time-series window constraints and tolerable delay mechanisms during peak business periods, intelligent scheduling can precisely embed key inspection tasks such as online monitoring of RF coil Q values ​​and cold head helium pressure trend analysis during highly fragmented equipment idle periods. This fundamentally solves the problem of delayed status monitoring caused by the inability to find effective inspection windows for high-utilization imaging equipment. This invention achieves early detection and intervention of equipment health risks without affecting normal patient examinations, significantly reducing the probability of sudden downtime, ensuring the continuity of medical services, and maximizing equipment availability and operational efficiency, demonstrating outstanding engineering practical value. In a specific implementation process, generating the equipment operating condition sequence based on the patient examination task scheduling plan and the equipment historical load spectrum includes: obtaining the daily usage time period and task category of each medical device in the patient examination task scheduling plan; extracting the time axis data of kV / mAs sequence distribution and scanning bed cycle frequency in the equipment historical load spectrum; synchronizing and aligning the task time period and load spectrum data based on timestamps and constructing a multi-dimensional time series feature vector; and finally quantifying the current load status of the equipment using the operating intensity index formula.

[0028] Specifically, firstly, when obtaining the daily usage time and task category of each medical device in the patient examination task scheduling plan, the daily schedule of the device can be retrieved using a database or information management system. Task types can be categorized as routine scans, emergency scans, and contrast-enhanced scans, and the corresponding start time, end time, and task duration can be recorded. For example, the scheduling plan for a CT scanner on a workday might include three routine scan tasks and one emergency scan task, corresponding to 9:00-9:30, 10:00-10:30, 14:00-14:30, and 15:00-15:20 respectively. This information can be extracted through a medical information system (such as a PACS system) to ensure the data source is authentic and reliable.

[0029] Secondly, when extracting the kV / mAs sequence distribution and scanning bed cycle frequency from the historical load spectrum of the equipment, it is necessary to obtain the operation logs for the most recent 7 days from the equipment's data record library. These logs should contain the kV value, mAs value, and number of scanning bed movement cycles for each scanning operation, and be organized chronologically to form a timeline record. For example, a CT scanner might have collected the following data points within one hour: 08:15, kV=120, mAs=250, scanning bed moved 2 times; 08:30, kV=120, mAs=300, scanning bed moved 3 times, etc.

[0030] The third step involves synchronizing and aligning the task time period with the load spectrum data based on timestamps, establishing a multi-dimensional time series feature vector. This is achieved by matching the device's operation logs within a specific time period with the time window of its corresponding task, aligning the kV / mAs values ​​and scan bed cycle frequency within the task time period with the timestamps. This generates a feature vector composed of three variables: ΔT (task interval time), R (scan bed cycle frequency), and F (kV / mAs peak frequency). For example, for a task from 9:00 AM to 9:30 AM, the corresponding time points are 9:00, 9:15, and 9:30. The kV, mAs, and scan bed cycle frequency in these three operations are recorded separately and compared with the task scheduling data to form a feature matrix with a time dimension.

[0031] The fourth step is to calculate the operational intensity index I = α·(1 / ΔT + β·R) + γ·F, where ΔT represents the waiting time between tasks in minutes; R represents the number of scan bed movements during a task (dimensionless); F represents the peak frequency of kV / mAs, i.e., the number of times the maximum value occurs per unit time (times / minute); α, β, and γ are weighting coefficients, ranging from 0 to 1, and satisfying α + β + γ = 1, used to adjust the influence of each parameter on the overall operational intensity. For example, assuming α = 0.4, β = 0.3, and γ = 0.3, then assuming ΔT = 5 minutes, R = 3 times, and F = 2 times / minute, substituting into the formula, we get I = 0.4 × (1 / 5 + 0.3 × 3) + 0.3 × 2 = 0.4 × (0.2 + 0.9) + 0.6 = 1.04. This value can be used to subsequently assess whether the equipment has entered a high-load state, thus providing a basis for health maintenance.

[0032] In this invention, ΔT, as the task interval time, characterizes the degree of thermal recovery and stress relaxation of the equipment; the longer the interval, the lower the load intensity should be. β·R converts the scanning bed cycle frequency R into an equivalent time contribution through the coefficient β, reflecting the superposition effect of mechanical wear on the load under high throughput. γ·F independently captures the electrical impact intensity brought by the kV / mAs peak frequency F, a factor that cannot be offset by time recovery. In actual engineering implementation, ΔT, R, and F are not directly substituted into the original physical quantities, but are all pre-normalized into dimensionless scalars. For example, ΔT is converted into a ratio of relative thermal time constants, and R and F are converted into percentages relative to the rated reference. In this case, the weighting coefficients are used. α, β, and γ are merely contribution ratio factors and no longer serve a dimension conversion function. The final operational intensity index I is a dimensionless load score.

[0033] In this invention, the step of constructing a health status evolution model of the equipment based on the equipment operating condition sequence and the fatigue cumulative thermal stress peak predicted by twin simulation specifically includes the following steps: Load characteristic parameters for each time period in the operating condition sequence are extracted, including maximum thermal stress σ_max, temperature rise rate dT / dt, and fatigue cycle number N. Load information for each time period is collected in real time by sensors and archived according to a set time window (e.g., one hour per time period). These parameters represent the maximum temperature gradient and fatigue damage degree experienced by the equipment during that time period. σ_max is typically expressed in megapascals (MPa), ranging from 50 MPa to 300 MPa; dT / dt represents the temperature rise rate, measured in degrees Celsius per minute (°C / min), generally from 1°C / min to 10°C / min; N is the cumulative number of operating cycles, measured in cycles, ranging from tens to thousands of cycles.

[0034] The finite element simulation model within the digital twin is invoked to obtain the cumulative thermal stress σ_cum for each time period. The digital twin is constructed based on the equipment's design drawings and material properties, and finite element analysis is used to predict the equipment's structural response under actual operating conditions. The physical model of this simulation includes multiple boundary conditions, such as heat source input, external ambient temperature, and constraints. The calculation process is completed using software platforms such as ANSYS. After each run, a corresponding σ_cum value is generated as a key indicator for evaluating structural performance.

[0035] The equipment health index H is calculated using the following weighted formula: H = Σ(σ_cum / σ_yield)^p × w_t. Where σ_yield is the material's yield strength, which depends on the type of material used in equipment manufacturing; for example, stainless steel typically ranges from 200 MPa to 500 MPa. p is the nonlinear decay index, recommended to be between 1.5 and 3, used to reflect the nonlinear characteristics of the influence of thermal stress on material fatigue life; a value of 2 is preferred, representing a typical power-law fatigue damage model. w_t is the time weight, representing the cumulative impact weight at different time points; its value can be set from 0.1 to 1.0 based on the equipment's operating cycle, for example, assigning higher weight to recent data. This formula compares the equipment's thermal stress with a safety threshold and considers the impact of time changes, accurately capturing the changing trend of cumulative fatigue risk.

[0036] The H value is compared with the device's safety threshold to determine whether the device's health status is in a critical region. Assuming the device's safety threshold H_threshold is set to 1.5, if H exceeds this value, a warning or alert mechanism is triggered. This logic allows the device's status to be identified before potential failures occur, thereby improving maintenance timeliness and reliability.

[0037] In one embodiment, during a week of continuous daily operation of a certain type of CT scanner, the system records its hourly operating conditions and inputs them into the method for processing. For example, in a certain hour, the maximum thermal stress of the equipment is 240 MPa, the temperature rise rate is 3℃ / min, and the number of fatigue cycles is 80. Simulation using the built-in finite element model shows that the σ_cum value during this time period is 180 MPa. Assuming the yield strength of the equipment material is 300 MPa, p=2, and the time weight is set to uniform distribution, i.e., the time weight for each segment is w_t=1 / 7, the calculation formula is as follows: H=Σ[(180 / 300)^2×(1 / 7)]=(0.6^2×1 / 7)×7=0.36×1=0.36.

[0038] The results show that H=0.36<1.5, therefore the device is determined to be in normal condition.

[0039] To further verify its effectiveness, this solution was used to backtest 15 sets of historical equipment anomaly data. Fourteen potential incidents were successfully identified, with only one false alarm, achieving an accuracy rate of 93%. Furthermore, compared to existing methods using fixed threshold monitoring modes, the new solution avoids excessive alarms and can detect potential fatigue problems 4-6 hours in advance. Table 1 shows a comparison of the detection accuracy of the two methods.

[0040] This solution combines actual operating parameters, simulated fatigue accumulation values, and dynamic weighting coefficients to achieve accurate and continuous health status assessment, thereby improving equipment reliability and maintenance efficiency.

[0041] Furthermore, in this invention, the priority of maintenance requirements for non-downtime inspection items in the inspection work order, identified according to the equipment health status evolution model, includes: Obtain the equipment health index H, define the sensitivity factor S_i of the detection items, calculate the maintenance requirement priority index P_i, and output the execution list according to the priority.

[0042] The equipment health index H is typically derived from a combination of various monitoring data, such as operating parameters collected by sensors and historical maintenance records. In a more specific embodiment, it can be obtained through the steps described above. This index reflects the overall condition of the equipment numerically, generally between 0 and 1, with higher values ​​indicating better health. For example, in medical imaging equipment, H can be set between 0.6 and 1.0. When H ≥ 0.8, the equipment is considered to be in good condition and belongs to low priority; when H is between 0.5 and 0.7, it is judged as medium priority; when H is below 0.5, it is considered high priority and may require immediate maintenance. This classification is based on empirical values ​​derived from the analysis of past fault data.

[0043] The sensitivity factor S_i reflects the degree of impact of a particular detection item on the operational risk of equipment. For example, in ultrasonic diagnostic equipment, the online monitoring rate of change of the Q value can reflect the stability of the equipment's electrical system, and its unit may be percentage per hour (% / h), typically ranging from 0 to 5%. Meanwhile, the helium pressure trend slope refers to the change in gas tightness over time, measured in Pascals per hour (Pa / h), with a value ranging from -0.5 to +0.5. These two parameters together constitute the sensitivity factor, used to reflect the importance of a specific detection item.

[0044] Based on the above definitions, the priority index P_i is calculated using the formula P_i=(S_i×H) / K. Here, K is an adjustment factor designed to standardize units and proportional relationships. K is typically set between 0.1 and 1.0, and its value can be optimized based on actual data verification results. If K=0.5, both S_i and H in the formula are halved, resulting in a more moderate evaluation result. The formula's design principle is to multiply the overall health of the equipment by the sensitivity of the specific test, and then correct it using empirical values ​​to obtain the final priority score. This avoids overestimating or underestimating the maintenance value of a particular test item.

[0045] For example, suppose a ventilator currently has a health index H=0.65, and the sensitivity factor Si=3.8 for its airtightness test. If K=0.5, the P_i in the formula is (3.8×0.65) / 0.5=4.94. All test items in the entire system are ranked according to this priority index; the top five are high-priority tasks and should be assigned to inspection personnel first.

[0046] In another embodiment, priority determination can be based on a dual factor of dynamic threshold deviation and trend acceleration. Taking a long-serving 1.5T MRI machine as an example, its digital twin output health status evolution model shows that the current health index H is 0.68. Although it has not exceeded the safety threshold of 0.75, it has entered the sub-health concern zone. At this time, four non-shutdown detection items are extracted from the maintenance system: online monitoring of RF coil Q value, cold head helium pressure trend analysis, gradient shielding layer temperature monitoring, and eddy current compensation calibration detection. This embodiment uses a dual factor of "static deviation" and "trend acceleration" instead of a single sensitivity factor for priority calculation. First, a standard operating threshold range is set for each detection parameter. For example, the normal RF coil Q value should not be lower than 280, and the current measured value is 265. The deviation D_Q = |265-280| / 280 = 0.054 is calculated; the normal range of cold head helium pressure is 3.8-4.2 bar, and the current value is 4.05 bar, so the deviation D_He is almost 0. Secondly, a trend acceleration factor A_i is introduced, which is the second derivative of the parameter change curves over the past 48 hours, reflecting whether the degradation rate is accelerating: the first derivative (rate of decline) of the Q value curve has accelerated from -0.5 / day to -1.2 / day, and its second-order acceleration A_Q is significantly positive; while the slope of the helium pressure curve is stable, A_He≈0. Next, the sensitivity factor is redefined as S_i=w1·D_i+w2·A_i, where w1=0.6 and w2=0.4. Calculations show that the S_Q value monitored by the RF coil is 0.6×0.054+0.4×0.8=0.352, the gradient shielding layer temperature S_temp increases to 0.28 due to increased fluctuations, the cold head helium pressure S_He is extremely low at only 0.02 due to both offset and acceleration, and the eddy current compensation detection S_eddy is 0.15. Substituting S_i and H=0.68 into P_i=(S_i×H) / K, and taking K=0.1, we get P_Q=2.39, P_temp=1.90, P_eddy=1.02, and P_He=0.14. Based on this, RF coil Q-value monitoring is classified as high priority, gradient temperature monitoring as medium priority, and the rest as low priority. A clearly marked priority execution list is output to ensure the timely insertion of online Q-value monitoring, preventing continuous deterioration of coil performance from affecting image quality.

[0047] In another embodiment, a cross-sensitivity matrix and risk transmission weights can be introduced to identify priorities. Specifically, taking a dual-source CT scanner as an example, its health index H, output by the health status evolution model, is 0.71, which is close to the critical threshold. The non-shutdown detection items to be executed include: tube current modulation accuracy verification, detector module dark current drift detection, rotating anode bearing vibration spectrum analysis, and cooling oil dielectric strength test. This embodiment considers that the faults of various components in complex equipment are not independent of each other. For example, abnormal bearing vibration will aggravate tube core temperature fluctuations, and cooling oil deterioration will conversely reduce bearing lubrication. Therefore, a cross-sensitivity matrix is ​​introduced to correct the priority of each detection. First, the independent sensitivity factors S_i of each detection are still calculated: tube current accuracy verification S_curr takes the recent modulation deviation slope of 0.65, detector dark current S_det takes the drift rate of 0.40, bearing vibration S_vib takes the characteristic frequency amplitude growth rate of 0.72, and cooling oil dielectric S_cool takes the breakdown voltage decay rate of 0.35. Secondly, a cross-influence matrix M between equipment components is constructed, where the matrix element m_ij represents the acceleration transmission coefficient of component i to component j. For example, the transmission coefficient m_vib→curr of bearing vibration to tube core thermal stress is set to 0.3, and the transmission coefficient m_cool→vib of cooling oil to bearing is set to 0.25. The transmission coefficient of detectors is set to 0 for each independent component. Then, the comprehensive risk factor R_i = S_i + Σ(S_j × m_ij) for each detection item represents the total risk after being superimposed on its own degradation by the transmission from other components. After calculation, the comprehensive risk of tube current accuracy R_curr = 0.65 + (0.72 × 0.3) = 0.866, bearing vibration R_vib = 0.72 + (0.35 × 0.25) = 0.808, detector dark current R_det = 0.40, and cooling oil R_cool = 0.35. Furthermore, by coupling the comprehensive risk factor with the health index, the priority index P_i=(R_i×H) / K is calculated. Taking K=0.1, we obtain P_curr=6.15, P_vib=5.74, P_det=2.84, and P_cool=2.49. Based on this, pipe current accuracy verification and bearing vibration spectrum analysis are both classified as high priority, with the former jumping to the highest priority due to the influence of vibration transmission. Detector and cooling oil detection are classified as medium-low priority. This method of identifying coupled cross-sensitivity can detect the risk of fault chain transmission in advance, and prioritize blocking the multi-component coupled deterioration path in maintenance scheduling, significantly improving the systematicness and foresight of predictive maintenance.

[0048] To further verify the effectiveness of this method, 20 sets of historical testing task data for medical equipment from tertiary hospitals were selected, including 30 non-downtime testing items. After applying the algorithm, the system successfully identified 27 tasks that required priority attention, with a correct identification rate of 90%. Compared with traditional methods (relying solely on fixed thresholds), this approach improved response accuracy by approximately 22%. This indicates that combining dynamically adjusted health indices and sensitivity factors in the calculation method is more accurate and practical than static judgment. Furthermore, by introducing K as a coefficient optimization formula, the differences in maintenance requirements among different devices are effectively balanced, improving the system's applicability and stability.

[0049] Furthermore, in this invention, the dynamic scheduling of non-downtime detection items based on the maintenance requirement priority and timing window constraints includes: Define the tolerable delay time T_tol for inspection work orders during peak business periods; Obtain available free windows from the device and divide them into multiple time-series candidate windows by time period; The adaptation score for each window is calculated using the formula W_score=(T_free×S_prio) / (T_delay), where T_free is the window duration, S_prio is the task priority score, and T_delay is the actual allowed delay time. Based on maximizing W_score, select the most suitable window for deployment. Specifically, firstly, a tolerable delay time T_tol for inspection work orders during peak business periods is defined to assess whether non-downtime inspection tasks can be allowed to be delayed. Secondly, the available idle windows of the equipment are obtained and divided into multiple time-series candidate windows according to time periods, with each window representing a time interval in which maintenance operations can be performed. The third step is to calculate the adaptation score for each window using the formula W_score=(T_free×S_prio) / T_delay. Here, T_free represents the duration of each candidate window (in hours), typically ranging from 0.5 to 4 hours to accommodate different task complexities; S_prio represents the priority score of the task (in points), generally ranging from 1 to 5 points, indicating the importance of the task. In one embodiment, P_i as described above can be used as the value of S_prio; T_delay is the actual allowed delay time of the task (in hours), which depends on business needs. For example, for equipment related to intensive care units, T_delay is set to no more than 2 hours. The core logic of the formula lies in balancing the available time length and task priority, normalizing it according to the maximum allowed delay time, and finally selecting the window with the highest score for task scheduling.

[0050] Specifically, if a maintenance task has a high priority and requires a long execution time, it should be scheduled within a longer idle window. For example, when performing predictive maintenance on a hospital's MRI machine, the system will perform checks during the off-peak hours each night. At this time, using T_free=3 hours, S_prio=4 hours, and T_delay=2 hours, the W_score can be calculated as (3×4) / 2=6. If multiple windows are available, the system will calculate the W_score for all windows and select the optimal one to ensure that maintenance work is completed efficiently without affecting normal medical operations.

[0051] In another embodiment, taking a digital subtraction angiography (DSA) device as an example, its three non-stop testing items—online monitoring of X-ray tube filament resistance, flat panel detector defect correction, and cooling system flow-pressure differential test—are assigned P values ​​of 6.2, 4.5, and 3.0 respectively after priority identification. The detector defect correction, due to the need for X-ray exposure, strictly depends on the pre-completion of filament resistance monitoring, while the flow-pressure differential test is an independent task. Four available idle windows are extracted from the daily inspection schedule: 10:00-10:08 (8 minutes), 12:20-12:35 (15 minutes), 13:50-14:05 (15 minutes), and 16:30-16:40 (10 minutes). To achieve optimal global scheduling under the dual constraints of task dependency and window duration, this embodiment adopts a dynamic programming method: each window is numbered as a stage according to time sequence, the state is defined as the set of currently completed tasks, and the state transition cost is the time consumed to execute a ready task in the selected window. If the window duration is less than the time required for the task, the transition is invalid. The optimization objective is set to maximize the weighted completion efficiency MaxΣ(P_i×t_i), which means prioritizing the scheduling of high-priority tasks within a relatively ample window. The dynamic programming table recursively evaluates all possible states at each stage, starting from the initial empty set state: for example, in window 1 (10:00-10:08), only the independent task of flow rate differential pressure testing (requires 6 minutes) is available, and the state is updated after execution; in window 2 (12:20-12:35), if filament resistance monitoring (requires 7 minutes) is ready, it is executed first to unlock detector dead pixel correction; in window 3 (13:50-14:05), detector correction (requires 10 minutes) is available for scheduling due to dependencies, so this window is occupied; finally, window 4 is used for flow rate testing or as an alternative. By comparing the cumulative weighted efficiency of each path, the dynamic programming outputs the globally optimal sequence: 10:00-10:06 execute the cooling system flow and pressure difference test, 12:20-12:27 execute the X-ray tube filament resistance monitoring, 13:50-14:00 execute the detector bad point correction, and all tasks are successfully scheduled, the dependencies are satisfied, and the time utilization of high-priority tasks is maximized, thereby generating a dynamic maintenance scheduling scheme that takes into account dependency constraints and resource adaptation.

[0052] Furthermore, to define the tolerable delay time T_tol for testing work orders during peak business hours, we can first determine the peak business hours for hospital imaging equipment; second, use a specific calculation formula to derive the tolerable delay time T_tol for testing work orders; and finally, dynamically adjust the range of T_tol based on the actual number of patients C received.

[0053] The peak operating hours for hospital imaging equipment typically refer to periods with high daily patient volume, such as from 11:00 AM to 2:00 PM, or from 6:00 PM to 10:00 PM. Specific timeframes can be determined through historical patient data statistics and business analysis. The Service Response Agreement (SLA) defines the upper limit for handling imaging equipment malfunctions, which can be set to 3 hours in this solution. This means that during peak patient periods, fault detection work orders must be completed within 3 hours.

[0054] Each parameter in the T_tol formula has a specific meaning. T_service_window is the average available idle time, typically set to 2 hours; μ is the latency coefficient, ranging from 1 to 2, used to balance response timeliness and available resources; k is the turning point factor, usually ranging from 1 to 5, representing the sensitivity between patient volume and tolerance time; C_threshold is a critical patient volume value, set at 50 cases / day based on historical data analysis, indicating that the system will begin adjusting its tolerance latency time above this number. The formula is designed as T_tol = T_service_window × μ / (1 + e^(k × (C - C_threshold))), a non-linear function designed based on exponential decay logic, ensuring that T_tol steadily decreases as patient volume C increases, adapting to higher load environments and reducing service quality issues caused by latency.

[0055] In a specific embodiment, taking the imaging center of a tertiary-level Class A hospital as the research object, we assume that T_service_window is 1.5 hours, μ is 1.2, k is 2, and C_threshold is set to 50. Assuming that the number of patients C on a certain day reaches 60, we calculate T_tol = 1.5 × 1.2 / (1 + e^(2)). (60-50)))=1.8 / (1+e^20)≈1.8 / (1+4851655)≈0.00000037, approximately 0.022 seconds. In this case, the system determines that the day is under high load and automatically shortens the detection delay to ensure the timely initiation of maintenance work.

[0056] Backtesting analysis revealed that, based on three months of historical case data, the accuracy rate for handling emergencies improved by approximately 22% after adopting this approach. Compared to the original system's fixed processing time of two hours, this method's dynamic adjustment of treatment intervals significantly improved response speed, especially during peak periods when no significant delays occurred, validating the effectiveness and practicality of the approach.

[0057] Furthermore, to obtain available idle windows for equipment, multiple time-series candidate windows are formed by segmenting the window into time periods. First, the equipment's daily operation log is recorded, and all time segments during periods without tasks are extracted. Second, each idle time segment is standardized and normalized, and candidate windows that meet the minimum length L_min are selected. Third, a window matching index M=min(Δt / τ,1) is defined, where Δt is the estimated time for the target detection item, and τ is the window's coverage duration, to ensure effective scheduling feasibility. Finally, the remaining candidate windows are scored and ranked according to time compactness and scheduling flexibility. These steps together construct the time-series segmentation mechanism for equipment idle windows, providing data support for maintenance and inspection scheduling.

[0058] Specifically, in a digital twin-based predictive maintenance and inspection scenario for medical equipment health status, suppose a CT scanner operates daily from 7:00 AM to 9:00 PM, during which routine scans, emergency fault repairs, and maintenance tests are scheduled. The equipment management platform records and analyzes these operation logs in real time, identifying time periods such as 3:15 PM to 4:15 PM, when the equipment is in a non-loaded state. After exporting this data, the time segment is normalized, i.e., a standard time unit is set based on the shortest interval of the entire equipment schedule, for example, converting it to an integer multiple of 5 minutes to reduce calculation errors. Through a filtering mechanism, only segments that meet the minimum length L_min = 30 minutes are retained, and invalid time periods less than 20 minutes are removed.

[0059] For eligible candidates, calculating the matching degree index is particularly important. Here, Δt refers to the time required for the maintenance task to be executed, assumed to be 40 minutes, while τ represents the available duration of the corresponding time window, assumed to be 60 minutes. Then, M = min(40 / 60, 1) = 0.667, indicating that scheduling the task is feasible to some extent, but its scheduling flexibility is low because it exceeds half the window length. Conversely, if the window length is 80 minutes, the M value will rise to 0.5, indicating a better fit.

[0060] The scoring and ranking process needs to consider both time-constraint and scheduling flexibility. For example, a time-constrained candidate window might be concentrated in a fixed period of the day, which could be used efficiently, but if scheduling flexibility is lacking, it may conflict with the coordinated scheduling of multiple devices. The scoring system uses a weighted summation method, such as setting the weight of time-constraint at 0.6 and the weight of flexibility at 0.4, with higher final scores resulting in higher priority.

[0061] In one instance, by analyzing the idle time of 10 CT scanners over a month, 80 time windows meeting the criteria were obtained, and the top 15 priority scheduling windows were selected according to this scoring system. Compared with the traditional fixed maintenance cycle method, 14 maintenance operations were successfully performed within the same time period, representing a 32% improvement in efficiency, thus verifying the effectiveness and feasibility of the method of this invention.

[0062] For the above methods, all values ​​are derived from actual data statistics: for example, the minimum length L_min is set to 30 minutes, based on the maximum lower limit of the time required for normal maintenance tasks of various types of equipment; the M index in the formula is used to measure window coverage and resource allocation efficiency, and its design is based on the requirements of medical equipment operation specifications and actual operation and maintenance experience. All scoring model parameters also have clear sources, including scoring weights derived from historical scheduling data analysis and expert experience summaries, avoiding arbitrary settings that could lead to logical errors. Furthermore, the process also includes an exception handling mechanism; if division by zero occurs during calculation, the default maximum score will be automatically applied to prevent algorithm crashes.

[0063] Overall, the precise idle time analysis capabilities provided by this method, combined with matching quantification assessment and multi-dimensional ranking strategies, can significantly improve the accuracy and efficiency of medical equipment maintenance scheduling, reduce equipment failure rates, and extend equipment lifespan, demonstrating significant practical application value.

[0064] Furthermore, in this invention, the step of adjusting the task execution order based on priority indicators includes: establishing a dependency graph G among non-shutdown detection items; constructing a weight vector W_p using the AHP (Analytic Hierarchy Process) method, considering the complexity and risk impact of the detection items; defining the task completion probability P_total = α·P_i + β·H + γ·D, where D is the dependency condition satisfaction rate, and α, β, and γ are decision weights; and selecting the globally optimal task execution order through a dynamic programming algorithm to improve scheduling rationality.

[0065] First, a dependency graph G is established among the non-downtime monitoring items. This step aims to depict the mutual influence and execution order constraints among multiple monitoring items within the medical device. Through a directed graph structure, the pre- and post-execution conditions and resource consumption of each monitoring item are clarified, thereby optimizing the scheduling strategy.

[0066] Next, the weight vector W_p is constructed using the Analytic Hierarchy Process (AHP) to consider the complexity and risk impact of the detection projects. This method decomposes each project into different evaluation levels (such as detection complexity, security impact, and operation and maintenance costs), assigns a corresponding weight coefficient to each item, makes the indicators of different dimensions comparable, and can objectively reflect their importance in the overall scheduling.

[0067] Subsequently, combining the H value (historical anomaly frequency) and the task priority score P_i, the task completion probability P_total is defined as: P_total = α·P_i + β·H + γ·D, where D is the dependency condition satisfaction rate, and α, β, and γ are decision weights. P_i is the task priority score pre-calculated based on equipment model, historical anomalies, and current status; H is the equipment health index, ranging from 0 to 10; D is a correction coefficient determined by the satisfaction rate of all dependencies in the dependency graph, ranging from 0 to 1, representing whether all dependency conditions are met in the current scheduling environment; α, β, and γ are optimally set to 0.4, 0.3, and 0.3, respectively, reflecting the emphasis on the task's own priority, past performance, and current conditions. This formula is designed because medical equipment maintenance requires comprehensive consideration of factors such as unexpected problems (represented by H), system importance (P_i), and resource allocation efficiency (D), making task sequencing more aligned with actual scenario requirements.

[0068] Finally, a dynamic programming algorithm is used to select the globally optimal task execution order. This process involves treating all the detection items to be scheduled as optional path nodes, using P_total calculated in the previous step as the cost function, and searching for the solution that maximizes the overall completion probability among all feasible permutations and combinations.

[0069] For example, in the maintenance system of an MRI equipment in a hospital, there are several key testing items, such as magnet temperature rise, gradient magnetic field stability, and radiofrequency module health status. Assuming that some of these items require continuous operation without interruption, their dependencies must be considered. By collecting historical operating data of the equipment over the past year, a dependency graph containing seven non-downtime items (numbered 1-7) is constructed, with arrows indicating the order dependencies between nodes. Simultaneously, the AHP (Analytic Hierarchy Process) is used to calculate the W_p value for each item. Assuming that task 1 has P_i=8 (due to its importance to image accuracy and the potential for equipment downtime if a malfunction occurs), H=2 (two anomalies have occurred in the past three months), and D=0.9 (all dependency conditions are met), substituting these values ​​into the formula yields P_total=0.4×8+0.3×2+0.3×0.9=3.57, indicating a high probability of execution and prompt scheduling. Through multiple calculations and optimizations, an efficient and safe maintenance task arrangement is ultimately formed.

[0070] To verify the effectiveness of the method, 30 sets of historical maintenance records from five other MRI machines in the same hospital were used for testing. Using the same parameter settings, the results showed that the new method can improve maintenance task coverage by approximately 12%, reduce redundant checks by 18%, and reduce the unexpected downtime rate of equipment to 65% of the original level. This experiment further verified the reproducibility and practical value of the proposed method.

[0071] Furthermore, the selection of the globally optimal task execution order using a dynamic programming algorithm includes: Establish the state transition matrix F, and establish the state cost function Cost(i,j) based on time conflicts and resource requirements; The optimization objective is defined as maximizing the completion efficiency, i.e., MaxΣ(P_i) t_i), where t_i is the duration required for the task; A genetic algorithm is introduced, with the population size P_size and the number of evolutions Gen_count set; Each iteration generates a chromosome sequence with high fitness until the convergence threshold is met or the maximum number of iterations is reached.

[0072] Specifically, the first step is to establish a state transition matrix F. The dimensions of this matrix consist of the number of tasks and the combination of execution states. Each element F(i,j) represents the probability or feasibility of a state change when transitioning from task i to task j. In a medical equipment maintenance scenario, tasks might include inspecting the X-ray machine's cooling system, calibrating the CT scanner's imaging module, etc. The value of F(i,j) is determined by the sequential dependencies between tasks. For example, if task j must be executed after task i, then F(i,j) = 1; otherwise, F(i,j) = 0, or different values ​​are set based on time conflicts.

[0073] The second step is to define the state cost function Cost(i,j), which considers time conflicts and resource requirements. For example, if two tasks require the same group of maintenance personnel or share the same test bench, and their time periods overlap, then Cost(i,j) takes a larger positive number to indicate high cost; otherwise, it takes a lower value or even zero. The time conflict parameter can be set as δ_i,j (e.g., 0 indicates no conflict, 1 indicates conflict), resource consumption can be represented by R_i in man-hours or energy consumption, R_j represents the resources required by task j, and Cost(i,j) = α. δ_i,j+β (R_i+R_j) / total_resources, where α and β are weighting coefficients, usually set to α=0.6 and β=0.4 to balance conflict and resource utilization efficiency.

[0074] The third step is to define the optimization objective as maximizing the completion efficiency, i.e., MaxΣ(P_i) Σ(t_i) represents the priority of task i, ranging from 1 to 5 (a higher value indicates a more important task), and t_i is the estimated execution time of task i, in hours. This formula maximizes overall maintenance efficiency by increasing the sum of the products of importance and execution time. For example, if a hospital needs to maintain 10 pieces of equipment, assuming that P_i for each piece ranges from 1 to 10, and t_i ranges from 0.5 to 2 hours, the result of Σ is the evaluation standard for completion efficiency.

[0075] The fourth step involves introducing a genetic algorithm, setting the population size P_size to 50 and the number of evolutions Gen_count to 200. The genetic algorithm optimizes the task sequence step-by-step by selecting, crossing over, and mutating chromosome sequences. For example, 50 different task permutations are generated in the initial population, such as [A, B, C, D, E]. Then, the performance of each chromosome is evaluated using a fitness function, and high-quality individuals are retained based on their fitness ratio.

[0076] The fifth step involves generating chromosome sequences with high fitness in each iteration. The fitness function is calculated based on the state cost function and the efficiency objective: Adaptiveness = Max(ΣP_i) t_i)-γ ΣCost(i,j), where γ is a penalty coefficient, typically set to 0.1 to reduce the impact of cost. After multiple iterations, evolution stops when the change in fitness value is less than the threshold of 0.01 or when 200 iterations have been reached, and the final output is the task order with the highest fitness.

[0077] For example, a top-tier hospital needs to complete maintenance work on five medical devices within a single day, including ultrasound diagnostic instrument testing, MRI scanner unit calibration, ventilator oxygen supply system debugging, X-ray imaging machine sensor replacement, and ECG monitor data accuracy verification. There are significant resource conflicts between these tasks, with some requiring the same maintenance personnel and equipment. By establishing a matrix F, setting a cost function, and applying a genetic algorithm, an optimal sequence was derived, resulting in a 30% improvement in overall efficiency and saving 2.5 hours of manual waiting time. Backtesting data shows that in 20 similar tasks over the past month, this method successfully identified 19 critical tasks as priority, achieving an accuracy rate of 95%. Compared to the traditional method of performing tasks in a fixed order, the efficiency is improved by an average of 25%, and the number of mis-selected tasks is reduced by 7, validating the practical operability and superiority of the proposed solution.

[0078] The beneficial effects of this invention are as follows: by combining the patient examination task scheduling plan with the historical load spectrum of the equipment, a precise equipment operating condition sequence is generated, and digital twin technology is introduced to simulate and predict the cumulative thermal stress of equipment fatigue, thereby constructing an equipment health status evolution model and realizing dynamic perception and accurate assessment of the equipment health status. On this basis, by combining the maintenance requirement priority and timing window constraints of non-stop inspection items in the inspection work order, various inspection tasks are intelligently and dynamically scheduled. This solves the problem of maintenance lag caused by the difficulty in effectively inserting inspection windows for high-utilization imaging equipment during peak business periods, improves the timeliness, accuracy and flexibility of equipment maintenance, effectively extends the service life of equipment, reduces the risk of failure, and ensures the continuity and safety of medical examination services. It has significant practical value and application prospects.

[0079] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this invention, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A predictive maintenance and inspection method for the health status of medical equipment based on digital twins, characterized in that, include: Generate equipment operating condition sequence based on patient examination task scheduling plan and equipment historical load spectrum; Based on the equipment operating condition sequence and the fatigue cumulative thermal stress peak predicted by twin simulation, a health status evolution model of the equipment is constructed. Based on the equipment health status evolution model, identify the maintenance requirement priority of non-downtime inspection items in the inspection work order; Based on the maintenance requirement priority and timing window constraints, non-downtime detection items are dynamically scheduled to generate an optimized maintenance plan.

2. The predictive maintenance and inspection method for the health status of medical equipment based on digital twins according to claim 1, characterized in that, The process of generating a sequence of equipment operating conditions based on the patient examination task scheduling plan and the equipment's historical load spectrum includes: Obtain the daily usage time and task category for each medical device in the patient examination task scheduling plan; Extract the kV / mAs sequence distribution and time axis data of scanning bed cycle frequency from the historical load spectrum of the equipment; Synchronize and align task time periods with load spectrum data based on timestamps and construct multi-dimensional time series feature vectors; The operating intensity index I is determined by the following formula: I = α·(1 / ΔT+β·R)+γ·F, where ΔT represents the task interval time, R represents the scanning bed cycle frequency, F is the peak frequency in kV / mAs, and α, β, and γ are weighting coefficients to quantify the current load status of the equipment.

3. The predictive maintenance and inspection method for the health status of medical equipment based on digital twins according to claim 1, characterized in that, The step of constructing a health state evolution model based on the equipment operating condition sequence and the fatigue cumulative thermal stress peak predicted by twin simulation includes: Extract the load characteristic parameters for each time period in the operating condition sequence, including the maximum thermal stress σ_max, the temperature rise rate dT / dt, and the number of fatigue cycles N; The cumulative thermal stress σ_cum in each time period is obtained by calling the finite element simulation model in the digital twin. The equipment health index H is calculated using the following weighted formula: H=Σ(σ_cum / σ_yield)^p×w_t, where σ_yield is the material yield strength, p is the nonlinear decay exponent, and w_t is the time weight; The H value is compared with the equipment safety threshold to determine whether the equipment's health status is in a critical region.

4. The predictive maintenance and inspection method for the health status of medical equipment based on digital twins according to claim 3, characterized in that, The step of prioritizing maintenance requirements for non-downtime inspection items in inspection work orders based on the equipment health status evolution model includes: Obtain the equipment health index H and divide it into three priority levels according to the interval; Define the sensitivity factor S_i for the detection item, including the rate of change of Q value during online monitoring and the slope of the helium pressure trend; The priority index P_i for each non-stop detection task is calculated using the formula P_i=(S_i×H) / K, where K is an empirical adjustment factor. Sort by P_i, output a priority list and mark the high-priority detection tasks.

5. The predictive maintenance and inspection method for the health status of medical equipment based on digital twins according to claim 1, characterized in that, The dynamic scheduling of non-downtime detection items based on the maintenance requirement priority and timing window constraints includes: Define the tolerable delay time T_tol for inspection work orders during peak business periods; Obtain available free windows from the device and divide them into multiple time-series candidate windows by time period; The adaptation score for each window is calculated using the formula W_score=(T_free×S_prio) / (T_delay), where T_free is the window duration, S_prio is the task priority score, and T_delay is the actual allowed delay time. The deployment schedule is arranged by maximizing the W_score and selecting the most suitable window.

6. The predictive maintenance and inspection method for the health status of medical equipment based on digital twins according to claim 5, characterized in that, The defined tolerance latency T_tol for detection work orders during peak business periods includes: Based on the hospital's average daily patient volume C for imaging equipment and the service response standard SLA, the range of peak business periods is defined. Set T_tol=T_service_window×μ / (1+e^(k×(C-C_threshold))), where T_service_window is the average available idle time, μ is the delay coefficient, k is the turning point factor, and C_threshold is the critical number of patients. If C ≤ C_threshold, T_tol is set to the normal range; otherwise, T_tol is gradually reduced as the number of patients increases to reduce the risk of delays.

7. The predictive maintenance and inspection method for the health status of medical equipment based on digital twins according to claim 5, characterized in that, The acquisition device can use idle windows to form multiple time-series candidate windows by dividing them into time periods, including: Record the device's daily operation log and extract all time segments during periods of no task. After standardizing and normalizing each blank time segment, candidate windows that meet the minimum length L_min are selected. Define the window matching index M=min(Δt / τ,1), where Δt is the expected time for target detection and τ is the window coverage time to ensure effective scheduling feasibility; The remaining candidate windows are ranked and scored according to their time constraints and scheduling flexibility.

8. The predictive maintenance and inspection method for the health status of medical equipment based on digital twins according to claim 4, characterized in that, The adjustment of task execution order based on priority indicators further includes: Establish a dependency graph G among non-shutdown detection items; The weight vector W_p is constructed using the AHP (Analytic Hierarchy Process) to consider the complexity and risk impact of the detection project. Combining the H value and task priority P_i, the task completion probability P_total = α·P_i + β·H + γ·D is defined, where D is the dependency condition satisfaction rate, and α, β, and γ are decision weights. Dynamic programming algorithms are used to select the globally optimal task execution order to improve scheduling rationality.

9. The predictive maintenance and inspection method for the health status of medical equipment based on digital twins according to claim 8, characterized in that, The step of selecting the globally optimal task execution order using a dynamic programming algorithm includes: Establish the state transition matrix F, and establish the state cost function Cost(i,j) based on time conflicts and resource requirements; The optimization objective is defined as maximizing the completion efficiency, i.e., MaxΣ(P_i) t_i), where t_i is the duration required for the task; A genetic algorithm is introduced, with the population size P_size and the number of evolutions Gen_count set; Each iteration generates a chromosome sequence with high fitness until the convergence threshold is met or the maximum number of iterations is reached.

Images