Fault detection and handling system and method based on a floor cleaning robot

Abstract

The application discloses a fault detection and processing system and method based on a floor cleaning robot, and relates to the technical field of artificial intelligence, and comprises the following steps: a perception acquisition unit collects robot operation data in real time; a fault detection unit processes the operation data to output an effective fault signal; a health parameter calculation unit calculates health parameters corresponding to three subsystems according to the effective fault signal; the floor cleaning robot system is divided into a navigation subsystem, an execution subsystem and a power supply subsystem; a fault type determination unit determines the fault type according to health parameters output by a navigation subsystem health parameter calculation module, an execution subsystem health parameter calculation module and a power supply subsystem health parameter calculation module in the health parameter calculation unit; and a processing unit executes a corresponding differentiated processing strategy according to the fault type output by the fault type determination unit. The application is a closed-loop technical scheme combining multi-dimensional fault detection, fault type determination and differentiated processing mechanisms.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a fault detection and handling system and method based on a floor cleaning robot. Background Technology

[0002] Floor cleaning robots are widely used in unattended scenarios such as shopping malls, airports, factories, and industrial parks. The equipment needs to operate continuously for a long time, so the requirements for the fault detection and handling mechanism of floor cleaning robots are very high.

[0003] Currently, fault detection and handling mechanisms for floor cleaning robots typically rely on simple threshold judgments or sensor status detection, lacking multi-sensor fusion, trend analysis, and correlation judgment. This leads to false alarms, missed alarms, and poor stability. Once a fault is triggered, a global shutdown and alarm are triggered. This simplistic and crude fault handling method has the following problems: minor faults in non-critical components can cause the entire machine to stop completely, reducing equipment online time and work efficiency; there is a lack of dynamic assessment and autonomous decision-making capabilities regarding the scope of fault impact. For example, when the cleaning actuator fails, the equipment still has complete navigation and re-charging capabilities, but traditional solutions directly paralyze it, unable to automatically return to the base station, severely reducing its availability in commercial scenarios; fault information is not shared between the navigation board, VCU, power board, and battery management system (BMS), and timestamps are not synchronized, making accurate diagnosis and traceability in the cloud impossible, resulting in low operation and maintenance efficiency. Summary of the Invention

[0004] This invention provides a fault detection and handling system and method based on a floor cleaning robot, which is a closed-loop technical solution that combines multi-dimensional fault detection, fault type determination, and differentiated handling mechanism.

[0005] According to one aspect of this disclosure, a fault detection and handling method based on a floor cleaning robot is provided, the method comprising the following steps: When the equipment is powered on, the navigation board, vehicle controller (VCU), and power board complete the timing start-up and time synchronization, and load the fault threshold, weight coefficient, and attenuation coefficient. The sensing and acquisition unit collects robot operation data in real time; The fault detection unit processes the operational data and outputs a valid fault signal. The health parameter calculation unit calculates the health parameters corresponding to the three subsystems based on the valid fault signals; the floor cleaning robot system is divided into three subsystems: navigation subsystem, execution subsystem, and power supply subsystem; The fault type determination unit determines the fault type based on the health parameters output by the navigation subsystem health parameter calculation module, the execution subsystem health parameter calculation module, and the power supply subsystem health parameter calculation module in the health parameter calculation unit. The processing unit executes the corresponding differentiated processing strategy according to the fault type output by the fault type determination unit; The interactive unit enables state linkage and cloud synchronization: the interactive unit updates the overall machine state machine, synchronizes the state to the UI and the cloud, and records fault logs.

[0006] In one possible implementation, the sensing and acquisition unit includes: a lidar, a ToF sensor, a binocular inertial module, an inertial measurement unit (IMU), a power controller, a temperature sensor, and a liquid level sensor, used to collect robot operation data in real time. All sensing data is sampled at a high frequency of 10Hz to 100Hz, filtered, and synchronized before being sent to the fault detection unit. LiDAR: Real-time acquisition of environmental point cloud, obstacle distance, and contour information for the detection of navigation and positioning anomalies, collision risks, and sensor communication anomalies; ToF sensor: Collects close-range 3D depth information for obstacle avoidance anomalies, main brush / squeegee jamming, and ground abrupt change detection; Binocular inertial module: Binocular cameras acquire visual features, image textures, and relative poses for visual loss, feature failure, and positioning drift detection; Inertial Measurement Unit (IMU): Collects triaxial acceleration, triaxial angular velocity, and attitude angle for detection of motion anomalies, rollover risk, and vibration exceeding limits; Power controller: Real-time acquisition of total current, branch current, bus voltage, individual unit voltage, and power for short circuit, overload, undervoltage, overvoltage, and abnormal power consumption detection; Temperature sensor: Collects the temperature of motor, drive board, power board, BMS, and navigation board for detection of overheating, abnormal heat dissipation, and component aging; Liquid level sensor: Collects the liquid level of the clean water tank and the wastewater tank for detection of water shortage, overflow, and abnormal replenishment.

[0007] In one possible implementation, the fault detection unit includes: an energy feature detection module, a prediction residual detection module, a correlation feature detection module, and a consistency detection module. Each module receives real-time data from the sensing and acquisition unit, extracts fault features from different dimensions, achieves multi-dimensional fusion judgment, and outputs a valid fault signal. The data sources for the energy characteristic detection module are the power controller, the walking motor driver, the cleaning actuator, the power board, and the battery management system (BMS). The energy characteristic detection module is used to collect electrical signals in real time and to determine abnormalities by calculating the accumulated energy of the sensor signals within a sliding time window. Calculate instantaneous energy: E instant (i) = X²(i) × R, where, E instant(i) represents the instantaneous energy at the i-th sampling moment; X(i) represents the real-time sampled value of current or voltage at the i-th sampling moment, which is a dynamic input quantity; R represents the impedance coefficient pre-calibrated by the system; Summing all instantaneous energies within a sliding time window of length L yields the total accumulated energy E within the time window. sum : , where t represents the number of the latest sample at the current time; Total accumulated energy E sum With preset threshold E th When E is compared, sum Greater than or equal to E th When an energy dimension anomaly is detected, it is used to identify hardware faults such as stall, short circuit, overload, and impact.

[0008] In one possible implementation, the prediction residual detection module uses data from LiDAR, ToF, binocular inertial module, IMU, power controller, and motor feedback as data sources. It obtains the predicted value based on the system operation model, calculates the difference between the predicted value and the actual collected value to obtain the single-moment residual, calculates the mean square error within the sliding time window, and determines that the prediction residual dimension is abnormal when the mean square error exceeds the preset threshold. Calculate the residual δ(i) at the current time: δ(i) = X pre (i) - X real (i); Among them, X pre (i) represents the predicted value output by the system model at time i, X real (i) represents the actual value collected by the sensor or hardware at time i; Calculate the mean square error of the residuals within the window: MSE = (1 / L)·Σδ²(i) Anomaly detection: Compare the total accumulated energy MSE with the preset threshold MSE. th Compare, when MSE is greater than or equal to MSE th When an anomaly is detected in the predicted residual dimension, it is used to identify potential faults related to progressive drift, component degradation, and performance decay in advance.

[0009] In one possible implementation, the correlation feature detection module uses the navigation board control commands and the actual feedback from the actuator as data sources. It first calculates the window mean of the command sequence and the feedback sequence, then calculates their covariance and standard deviation, and finally obtains the correlation coefficient. Under normal operating conditions, the commands and feedback are highly correlated, and the correlation coefficient approaches 1. When the actuator is out of control, communication packets are lost, transmission fails, or the motor is stuck, the correlation coefficient decreases significantly. When the correlation coefficient is lower than the threshold, it is determined that the correlation dimension is abnormal. Calculate the mean within the window: , , Where U represents the control command sequence. L represents the average value of the instruction sequence within the window, and the sliding time window length is L; Y represents the actual feedback sequence of the actuator. This represents the average value of the feedback sequence within the window; control commands include, for example, speed and current commands. The correlation coefficient between the control command and the execution feedback signal is calculated: Under normal operating conditions, the command and feedback are highly correlated, and the correlation coefficient is close to 1; when the actuator is out of control, communication packets are lost, the motor does not respond, or the transmission fails, the correlation coefficient decreases significantly. First, calculate the covariance Cov(U,Y): , Calculate the correlation coefficient Corr between the control command and the execution feedback signal: , where σ U σ represents the standard deviation of the instruction sequence. Y This represents the standard deviation of the feedback sequence; Finally, an abnormal correlation assessment is performed: when the absolute value of the correlation coefficient, |Corr|, is less than or equal to the threshold Corr... th When this occurs, it is determined to be an anomaly in the relevance dimension.

[0010] In one possible implementation, the consistency detection module performs cross-validation on the operating status of the navigation subsystem, execution subsystem, and power subsystem, and calculates the state deviation between each subsystem and the system consistency index. During normal operation, the state logic of each domain is consistent, and the consistency index approaches 1. If there is a mismatch between navigation and execution or a contradiction between the states of modules, the consistency index drops, and if it falls below the threshold, it is judged as an inter-domain consistency anomaly. Calculate the inter-subsystem deviation Δpq: Δpq = |Sp- Sq|, where Sp represents the state value of subsystem p and Sq represents the state value of subsystem q; subsystem p and subsystem q can be any two of the navigation subsystem, execution subsystem, and power subsystem. Calculate the consistency index Ccon: Ccon = 1 - (1 / N)·Σ(Δpq) Finally, a consistency anomaly is determined: when the consistency index Ccon is less than or equal to the threshold Cth, an inter-domain consistency anomaly is determined.

[0011] In one possible implementation, the effective fault signal is determined by: statistically analyzing the percentage of abnormal signals occurring within a unit of time; only when the percentage of abnormal time exceeds a set threshold is the multi-dimensional fusion judgment result confirmed as an effective fault, and instantaneous interference and single-point anomalies are completely filtered out. Calculate the percentage of abnormal time T prop : T prop = T abnormal / T total , among which, T prop This represents the duration T of the abnormal state within a given observation period. abnormal The percentage is used to determine whether the fault is a true, continuous fault or a transient disturbance; its value ranges from [0,1]. total Indicates the total duration of the observation window; Valid fault signal determination: When T prop Greater than or equal to the threshold T th When T is active, it is considered a valid fault signal; when T is active, it is considered a valid fault signal. prop Less than threshold T th When this is detected as transient interference, the current signal is filtered out. th Take a value of 0.3 to 0.5.

[0012] In one possible implementation, the health parameters of each subsystem are calculated using the following formula: , among which, S health : The current subsystem's health parameter, ranging from (0, 1]; ε: Health parameter decay coefficient, used to adjust the failure impact rate; W i : The severity weight of the i-th type of fault, pre-configured, with a value range of 0.2 to 1.0; N i T represents the cumulative number of times the i-th type of fault is triggered. i Fault duration factor; N max : The maximum permissible number of faults in the current subsystem; K s System stability coefficient; K t Time decay coefficient, used to reduce the weight of historical faults; Health parameter range: S health ∈[0.8, 1]: Normal; S health ∈[0.2, 0.8): Sub-health; S health ∈[0, 0.2): Functionality disabled; The methods for determining the fault type include: when the S of the power supply subsystem health < 0.2, or, the S of the navigation subsystem health < 0.2 and the execution subsystem's S healthIf the value is less than 0.2, it is determined to be fault type 1, indicating complete failure. When the navigation subsystem S health < 0.2 and the S of the power supply subsystem health ≥ 0.2 and the S of the execution subsystem health If the value is ≥ 0.2, it is determined to be fault type 2, and navigation fails; When executing subsystem S health < 0.2 and the S of the power supply subsystem health ≥ 0.2 and the S of the navigation subsystem health If the value is ≥ 0.2, it is determined to be a failure of type 3, execution failure. Fault type locking: When a series of faults reach a threshold within a short period of time, the fault type is locked to prevent frequent switching; it is unlocked after a period of continuous normal operation or after a power-on self-test.

[0013] In one possible implementation, the processing unit executes a corresponding differentiated processing strategy based on the fault type output by the fault type determination unit; specifically as follows: according to S health If the condition is determined to be sub-healthy, then an early warning, reporting, and continued operation will be implemented. If the fault type is determined to be 3, the cleaning mechanism will stop and automatic retraction will be triggered. If the fault is determined to be type 2, then automatic mode will be turned off and manual mode will be turned on. If the fault is determined to be type 1, then immediately activate emergency stop, power-off protection, and report a fatal fault.

[0014] The fault detection and handling system based on the floor cleaning robot includes: a sensing and acquisition unit, a fault detection unit, a health parameter calculation unit, a fault type determination unit, a processing unit, and an interaction unit. The sensing and acquisition unit includes: lidar, ToF, binocular inertial module, IMU, power controller, temperature sensor, and liquid level sensor; the power controller is used to acquire current and voltage in real time. The fault detection unit includes: an energy characteristic detection module, a prediction residual detection module, a correlation characteristic detection module, and an inter-domain consistency detection module; The health parameter calculation unit includes: a navigation subsystem health parameter calculation module, an execution subsystem health parameter calculation module, and a power supply subsystem health parameter calculation module; Fault type determination unit: Based on the health parameters output by the health parameter calculation unit, output fault type 1, fault type 2, and fault type 3; Processing Unit: Executes the corresponding processing strategy based on the fault type output by the fault type determination unit; The interaction unit includes: navigation board, VCU, power board, BMS, cloud server, and state machine.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This application employs a parallel detection method combining energy characteristics, prediction residuals, correlation characteristics, and inter-domain consistency characteristics, along with a time occupancy confirmation mechanism. It only responds to genuine faults that are persistent, have clear physical meaning, and are verifiable across multiple dimensions. This effectively filters out non-fault-related interference signals such as transient electromagnetic interference, signal spikes, communication jitter, and sensor noise. Compared to traditional schemes that rely solely on single-point threshold judgment, this significantly reduces the false alarm rate, resulting in a more stable and reliable system. It can operate stably for extended periods in complex electromagnetic and operational environments such as shopping malls, factories, and airports, without frequent false triggering of protection mechanisms or abnormal shutdowns due to interference.

[0016] Traditional floor cleaning robots shut down entirely when any module malfunctions, forcing shutdowns in numerous scenarios where degraded, self-healing, or locally shielded operations are possible, severely reducing continuous operation capability. This invention addresses this by employing independent three-domain health assessments and precise three-level fault classification. It differentiates the scope and severity of faults, implementing protective shutdowns only in cases of complete failure. In scenarios of sub-optimal health, partial failure, or non-critical component malfunctions, the equipment continues operation or degrades. This significantly reduces unnecessary downtime for floor cleaning robots, substantially improving their continuous operation capability and efficiency in unattended scenarios.

[0017] In the event of a failure in the execution subsystem but normal operation of the navigation and power subsystems, cleaning operations are automatically disabled while retaining full navigation, positioning, path planning, and motion control capabilities. The system can autonomously plan the optimal path back to the base station or a safe area. Even in the event of an actuator failure, automatic return to the charging dock is still possible, enhancing equipment safety and ease of maintenance. Conversely, in the event of a failure in the navigation subsystem but normal operation of the execution and power subsystems, automatic tasks and autonomous navigation functions are automatically disabled, while manual control and push-assist functions such as walking, cleaning, lighting, and audio are retained, significantly improving emergency response capabilities and user-friendly operation in case of equipment failure.

[0018] Health parameters change dynamically and smoothly with the accumulation and duration of faults and system stability, without jumps or fluctuations, enabling early warning of faults and facilitating preventative maintenance by users. Simultaneously, the system records the fault occurrence time, fault type, health status change curve, equipment position, task status, and environmental information in real time, synchronizing this data to the cloud, ensuring that the entire fault process is recordable, searchable, and traceable. Furthermore, it achieves integrated collaborative control of faults, status, tasks, power supply, and the cloud, enabling the entire machine to possess intelligent operation capabilities for self-diagnosis, self-decision-making, and self-handling, significantly improving the automation level and safety of the floor cleaning robot. Attached Figure Description

[0019] Figure 1A block diagram of a fault detection and handling system based on a floor cleaning robot according to an embodiment of the present disclosure is shown.

[0020] Figure 2 A flowchart of a fault detection and handling system and method based on a floor cleaning robot according to an embodiment of the present disclosure is shown. Detailed Implementation

[0021] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0022] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0023] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0024] Step 1: The device is powered on, and the navigation board, VCU, and power board complete the timing startup and time synchronization, and load parameters such as fault threshold, weight coefficient, and attenuation coefficient.

[0025] Step 2: The sensing and acquisition unit includes: LiDAR, ToF sensor, binocular inertial module, inertial measurement unit (IMU), power controller, temperature sensor, and liquid level sensor. It is used to collect robot operation data in real time, providing raw input for fault detection. All sensing data is sampled at a high frequency of 10Hz to 100Hz, filtered, and synchronized before being sent to the fault detection unit.

[0026] (1) LiDAR: Real-time acquisition of environmental point cloud, obstacle distance and contour information, used for navigation and positioning anomalies, collision risks and sensor communication anomalies detection; (2) ToF: Collects close-range three-dimensional depth information for obstacle avoidance anomalies, main brush / squeegee jamming, and ground change detection; (3) Binocular inertial module: Binocular camera collects visual features, image texture and relative pose for visual loss, feature failure and positioning drift detection; Inertial measurement unit (IMU): Collects three-axis acceleration, three-axis angular velocity and attitude angle for motion abnormality, rollover risk and vibration over-limit detection; (4) Power controller: Real-time acquisition of total current, branch current, bus voltage, individual voltage and power, used for short circuit, overload, undervoltage, overvoltage and power consumption abnormal detection; (5) Temperature sensor: collects the temperature of motor, drive board, power board, BMS and navigation board for detection of overheating, abnormal heat dissipation and device aging; (6) Liquid level sensor: collects the liquid level of the clean water tank and the liquid level of the sewage tank for detection of water shortage, overflow, and abnormal replenishment.

[0027] Step 3: The fault detection unit includes: an energy feature detection module, a prediction residual detection module, a correlation feature detection module, and a consistency detection module. Each of these modules receives real-time data from the sensing and acquisition unit, extracts fault features from different dimensions, achieves multi-dimensional fusion judgment, and outputs a valid fault signal.

[0028] (1) Energy Feature Detection Module The energy characteristic detection module receives data from the power controller, walking motor driver, cleaning actuator, power board, BMS, etc., and is used to collect electrical signals such as current and voltage in real time. Anomalies are identified by calculating the accumulated energy of the sensor signals within a sliding time window.

[0029] Calculate instantaneous energy: E instant (i) = X²(i) × R, where, E instant (i) represents the instantaneous energy at the i-th sampling moment; X(i) represents the real-time sampled value of current or voltage at the i-th sampling moment, which is a dynamic input quantity; R represents the impedance coefficient pre-calibrated by the system; Summing all instantaneous energies within a sliding time window of length L yields the total accumulated energy E within the time window. sum : , where t represents the number of the latest sample at the current time; Total accumulated energy E sum With preset threshold E th When E is compared, sum Greater than or equal to E th When an energy dimension anomaly is detected, it is used to identify hardware faults such as stall, short circuit, overload, and impact.

[0030] (2) Predictive residual detection module The predictive residual detection module uses data from LiDAR, ToF, binocular inertial modules, IMU, power controllers, motors, etc., as data sources. Based on the system operation model, it obtains predicted values, calculates the difference between the predicted values ​​and the actual collected values ​​to obtain the single-time residual, and calculates the mean square error within a sliding time window. When the mean square error exceeds a preset threshold, it is determined that the predictive residual dimension is abnormal.

[0031] Calculate the residual δ(i) at the current time: δ(i) = X pre (i) - X real (i), where X pre (i) represents the predicted value output by the system model at time i, X real (i) represents the actual value collected by the sensor / hardware at time i.

[0032] Calculate the mean square error of the residuals within the window (MSE): MSE = (1 / L)·Σδ²(i) Anomaly detection: Compare the total accumulated energy MSE with the preset threshold MSE th Compare, when MSE is greater than or equal to MSE th When an anomaly is detected in the predicted residual dimension, it is used to identify potential faults related to progressive drift, component degradation, and performance decay in advance.

[0033] (3) Correlation feature detection module The correlation feature detection module uses navigation board control commands and actual feedback from the actuator as data sources. It first calculates the window mean of the command sequence and feedback sequence, then calculates their covariance and standard deviation to obtain the correlation coefficient. Under normal operating conditions, commands and feedback are highly correlated, with the correlation coefficient approaching 1. When the actuator malfunctions, communication packets are lost, transmission fails, or the motor jams, the correlation coefficient decreases significantly. When the correlation coefficient falls below a threshold, it is determined to be an abnormal correlation dimension.

[0034] Calculate the mean within the window , , Where U represents a sequence of control commands (such as speed or current commands). represents the average value of the instruction sequence within the window, with a sliding time window length of L. Y represents the actual feedback sequence of the actuator. This represents the average value of the feedback sequence within the window.

[0035] Calculate the correlation coefficient between control commands and execution feedback signals. Under normal operating conditions, the commands and feedback are highly correlated, with the correlation coefficient approaching 1. When the actuator malfunctions, communication packets are lost, the motor does not respond, or the transmission fails, the correlation coefficient decreases significantly. First, calculate the covariance Cov(U,Y): , Calculate the correlation coefficient Corr between the control command and the execution feedback signal: , where σ U σ represents the standard deviation of the instruction sequence. Y This represents the standard deviation of the feedback sequence.

[0036] Finally, an abnormal correlation assessment is performed: when the absolute value of the correlation coefficient, |Corr|, is less than or equal to the threshold Corr... th When this occurs, it is determined to be an anomaly in the relevance dimension.

[0037] (4) Consistency Detection Module Cross-validation is performed on the operational status of the navigation subsystem, execution subsystem, and power supply subsystem to calculate the state deviation between each subsystem and the system consistency index. During normal operation, the state logic of each domain is consistent, and the consistency index approaches 1. If a mismatch occurs between navigation and execution, or a state contradiction occurs between modules, the consistency index decreases. If it falls below the threshold, it is judged as an inter-domain consistency anomaly.

[0038] Calculate the inter-subsystem deviation Δpq: Δpq = |Sp- Sq|, where Sp represents the state value of subsystem p and Sq represents the state value of subsystem q; subsystem p and subsystem q are any two of the navigation subsystem, execution subsystem, and power supply subsystem.

[0039] Calculate the consistency index Ccon: Ccon = 1 - (1 / N)·Σ(Δpq) Finally, a consistency anomaly is determined: when the consistency index Ccon is less than or equal to the threshold Cth, an inter-domain consistency anomaly is determined.

[0040] (5) Determination of valid fault signals: The percentage of continuous abnormal signals within a unit of time is statistically analyzed. Only when the percentage of abnormal time exceeds a set threshold is the multi-dimensional fusion judgment result confirmed as a valid fault, and instantaneous interference and single-point anomalies are completely filtered out.

[0041] Calculate the percentage of abnormal time T prop : T prop = T abnormal / T total , among which, Tprop This represents the duration T of the abnormal state within a given observation period. abnormal The percentage is used to determine whether the fault is a true, continuous fault or a transient disturbance; its value ranges from [0,1]. T total This indicates the total duration of the observation window.

[0042] Valid fault signal determination: When T prop Greater than or equal to the threshold T th When T (usually taken as 0.3 to 0.5) is considered a valid fault signal; when T prop Less than threshold T th If the signal is detected as transient interference, the current signal will be filtered out.

[0043] Step 4: Output the valid fault signal from the fault detection unit to the health parameter calculation unit to calculate the health parameters for each of the three subsystems (the floor cleaning robot system is divided into three subsystems: navigation subsystem, execution subsystem, and power supply subsystem). The fault detection unit updates the health parameters for the three subsystems every 100ms and outputs them to the fault classification unit.

[0044] Navigation subsystem: involves SLAM, relocalization, sensors, path planning, etc. Execution subsystem: involves walking, main brush, water pump, squeegee, valves, lights, audio, etc.; Power subsystem: This includes power board, BMS, battery, charging, power supply protection, etc.

[0045] For each of the three subsystems mentioned above, the health parameters of that subsystem are calculated using the following formula: , Among them, S health : The health parameter of the current subsystem, with a value range of (0, 1]. ε: Health parameter decay coefficient, used to adjust the rate of failure impact; W i The severity weight of the i-th type of fault is pre-configured and ranges from 0.2 to 1.0. N i : The cumulative number of times the i-th type of fault is triggered; T i Fault duration factor; N max : The maximum allowable number of faults in the current subsystem; K s System stability coefficient; K t Time decay coefficient, used to reduce the weight of historical faults.

[0046] Health parameter range: S health ∈[0.8, 1]: Normal S health ∈[0.2, 0.8): Sub-health S health ∈[0, 0.2): Functionality disabled Step 5: The fault type determination unit determines the fault type based on the health parameters output by the navigation subsystem health parameter calculation module, execution subsystem health parameter calculation module, and power supply subsystem health parameter calculation module in the health parameter calculation unit. The specific determination method is as follows: (1) When the S of the power supply subsystem health < 0.2, or, S of the navigation subsystem health < 0.2 and the execution subsystem's S health < 0.2; Then it is determined to be fault type 1 (complete failure); (2) When the S of the navigation subsystem health < 0.2 and the S of the power supply subsystem health ≥ 0.2 and the S of the execution subsystem health ≥ 0.2 The fault type is then determined to be 2 (navigation failure). (3) When the execution subsystem S health < 0.2 and the S of the power supply subsystem health ≥ 0.2 and the S of the navigation subsystem health ≥ 0.2 The fault type is then determined to be 3 (execution failure).

[0047] Fault type locked: If a series of consecutive failures reach a threshold within a short period of time, the failure type is locked to prevent frequent switching. Unlocked after continuous normal operation for a period of time or after power-on self-test passes.

[0048] Step 6: The processing unit executes the corresponding differentiated processing strategy based on the fault type output by the fault type determination unit; the details are as follows: If Shealth determines that a person is in a sub-healthy state, then an alert, a report, and continued operation will be initiated. If the fault type is determined to be 3, the cleaning mechanism will stop and automatic retraction will be triggered. If the fault is determined to be type 2, then automatic mode will be turned off and manual mode will be turned on. If the fault is determined to be type 1, then immediately activate emergency stop, power-off protection, and report a fatal fault.

[0049] Step 7: The interactive unit realizes state linkage and synchronization with the cloud.

[0050] The interaction unit updates the overall machine state machine, synchronizes the state to the UI and the cloud, and records fault logs, etc.

[0051] A system and method for fault detection, fault type determination, and differentiated handling of floor cleaning robots, achieving: Highly robust, low false alarm multidimensional fusion fault detection; An exponential decay model is introduced to calculate the health parameters of the navigation subsystem, execution subsystem, and power subsystem. Automatic fault type determination that strictly matches the operating logic of the floor scrubber; Based on the fault type, execute self-healing, downgrade, task disabling, manual control protection, pile return protection, and emergency shutdown; The fault is deeply linked to the overall machine status, power-on / off status, cloud, and base station.

[0052] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A fault detection and handling method based on a floor cleaning robot, characterized in that, The method includes the following steps: When the equipment is powered on, the navigation board, vehicle controller (VCU), and power board complete the timing start-up and time synchronization, and load the fault threshold, weight coefficient, and attenuation coefficient. The sensing and acquisition unit collects robot operation data in real time; The fault detection unit processes the operational data and outputs a valid fault signal. The health parameter calculation unit calculates the health parameters corresponding to the three subsystems based on the valid fault signals. The floor cleaning robot system is divided into three subsystems: navigation subsystem, execution subsystem, and power supply subsystem; The fault type determination unit determines the fault type based on the health parameters output by the navigation subsystem health parameter calculation module, the execution subsystem health parameter calculation module, and the power supply subsystem health parameter calculation module in the health parameter calculation unit. The processing unit executes the corresponding differentiated processing strategy according to the fault type output by the fault type determination unit; The interaction unit enables state linkage and cloud synchronization: the interaction unit updates the overall machine state machine, synchronizes the state to the UI and the cloud, and records fault logs.

2. The fault detection and handling method based on a floor cleaning robot according to claim 1, characterized in that, The sensing and acquisition unit includes: lidar, ToF sensor, binocular inertial module, inertial measurement unit (IMU), power controller, temperature sensor and liquid level sensor, used to collect robot operation data in real time. All sensing data is sampled at a high frequency of 10Hz to 100Hz, filtered and synchronized, and then sent to the fault detection unit. LiDAR: Real-time acquisition of environmental point cloud, obstacle distance, and contour information for the detection of navigation and positioning anomalies, collision risks, and sensor communication anomalies; ToF sensor: Collects close-range 3D depth information for obstacle avoidance anomalies, main brush / squeegee jamming, and ground abrupt change detection; Binocular inertial module: Binocular cameras acquire visual features, image textures, and relative poses for the detection of visual loss, feature failure, and positioning drift; Inertial Measurement Unit (IMU): Collects triaxial acceleration, triaxial angular velocity, and attitude angle for detection of motion anomalies, rollover risk, and vibration exceeding limits; Power controller: Real-time acquisition of total current, branch current, bus voltage, individual unit voltage, and power for short circuit, overload, undervoltage, overvoltage, and abnormal power consumption detection; Temperature sensor: Collects the temperature of motor, drive board, power board, BMS, and navigation board for detection of overheating, abnormal heat dissipation, and component aging; Liquid level sensor: Collects the liquid level of the clean water tank and the wastewater tank for detection of water shortage, overflow, and abnormal replenishment.

3. The fault detection and handling method based on a floor cleaning robot according to claim 1, characterized in that, The fault detection unit includes: an energy feature detection module, a prediction residual detection module, a correlation feature detection module, and a consistency detection module. Each module receives real-time data from the sensing and acquisition unit, extracts fault features from different dimensions, achieves multi-dimensional fusion judgment, and outputs a valid fault signal. The data sources for the energy characteristic detection module are the power controller, the walking motor driver, the cleaning actuator, the power board, and the battery management system (BMS). The energy characteristic detection module is used to collect electrical signals in real time and to determine abnormalities by calculating the accumulated energy of the sensor signals within a sliding time window. Calculate instantaneous energy: E instant (i) = X²(i) × R, where, E instant (i) represents the instantaneous energy at the i-th sampling moment; X(i) represents the real-time sampled value of current or voltage at the i-th sampling moment, which is a dynamic input quantity; R represents the impedance coefficient pre-calibrated by the system; Summing all instantaneous energies within a sliding time window of length L yields the total accumulated energy E within the time window. sum : , where t represents the number of the latest sample at the current time; Total accumulated energy E sum With preset threshold E th When E is compared, sum Greater than or equal to E th When an energy dimension anomaly is detected, it is used to identify hardware faults such as stall, short circuit, overload, and impact.

4. The fault detection and handling method based on a floor cleaning robot according to claim 3, characterized in that, The prediction residual detection module uses data from lidar, ToF, binocular inertial module, IMU, power controller, and motor feedback as data sources. It obtains predicted values ​​based on the system operation model, calculates the difference between the predicted values ​​and the actual collected values ​​to obtain the single-moment residual, and calculates the mean square error within the sliding time window. When the mean square error exceeds the preset threshold, it is determined that the prediction residual dimension is abnormal. Calculate the residual δ(i) at the current time: δ(i) = X pre (i)-X real (i); Among them, X pre (i) represents the predicted value output by the system model at time i, X real (i) represents the actual value collected by the sensor or hardware at time i; Calculate the mean square error of the residuals within the window: MSE = (1 / L)·Σδ²(i) Anomaly detection: Compare the total accumulated energy MSE with the preset threshold MSE. th Compare when MSE is greater than or equal to MSE. th When an anomaly is detected in the predicted residual dimension, it is used to identify potential faults related to progressive drift, component degradation, and performance decay in advance.

5. The fault detection and handling method based on a floor cleaning robot according to claim 3, characterized in that, The correlation feature detection module uses the navigation board control commands and the actual feedback from the actuator as data sources. It first calculates the window mean of the command sequence and the feedback sequence, then calculates their covariance and standard deviation, and finally obtains the correlation coefficient. Under normal working conditions, the commands and feedback are highly correlated, and the correlation coefficient is close to 1. When the actuator is out of control, communication packets are lost, transmission fails, or the motor is stuck, the correlation coefficient decreases significantly. When the correlation coefficient is lower than the threshold, it is judged as an abnormal correlation dimension. Calculate the mean within the window: ， ， Where U represents the control command sequence. The sliding time window represents the average value of the instruction sequence within the window, with a window length of L; Y represents the actual feedback sequence of the actuator. This represents the average value of the feedback sequence within the window; control commands include, for example, speed and current commands. The correlation coefficient between the control command and the execution feedback signal is calculated: Under normal operating conditions, the command and feedback are highly correlated, and the correlation coefficient approaches 1; when the actuator is out of control, communication packets are lost, the motor does not respond, or the transmission fails, the correlation coefficient decreases significantly. First, calculate the covariance Cov(U,Y): ， Calculate the correlation coefficient Corr between the control command and the execution feedback signal: , where σ U σ represents the standard deviation of the instruction sequence. Y This represents the standard deviation of the feedback sequence; Finally, an abnormal correlation assessment is performed: when the absolute value of the correlation coefficient, |Corr|, is less than or equal to the threshold Corr... th When this occurs, it is determined to be an anomaly in the relevance dimension.

6. The fault detection and handling method based on a floor cleaning robot according to claim 3, characterized in that, Consistency Detection Module: Performs cross-validation on the operating status of the navigation subsystem, execution subsystem, and power subsystem, and calculates the state deviation between each subsystem and the system consistency index. During normal operation, the state logic of each domain is consistent, and the consistency index approaches 1. If there is a mismatch between navigation and execution or a contradiction between the states of modules, the consistency index will decrease. If it falls below the threshold, it is judged as an inter-domain consistency anomaly. Calculate the inter-subsystem deviation Δpq: Δpq = |Sp- Sq|, where Sp represents the state value of subsystem p and Sq represents the state value of subsystem q; subsystem p and subsystem q can be any two of the navigation subsystem, execution subsystem, and power subsystem. Calculate the consistency index Ccon: Ccon = 1 - (1 / N)·Σ(Δpq) Finally, a consistency anomaly is determined: when the consistency index Ccon is less than or equal to the threshold Cth, an inter-domain consistency anomaly is determined.

7. The fault detection and handling method based on a floor cleaning robot according to claim 3, characterized in that, Valid fault signal determination: The percentage of continuous abnormal signals within a unit of time is statistically analyzed. Only when the percentage of abnormal time exceeds a set threshold is the multi-dimensional fusion judgment result confirmed as a valid fault. Instantaneous interference and single-point anomalies are completely filtered out. Calculate the percentage of abnormal time T prop : T prop = T abnormal / T total , among which, T prop This represents the duration T of the abnormal state within a given observation period. abnormal The percentage is used to determine whether the fault is a true, continuous fault or a transient disturbance; its value ranges from [0,1]. total Indicates the total duration of the observation window; Valid fault signal determination: When T prop Greater than or equal to the threshold T th When T is active, it is considered a valid fault signal; when T is active, it is considered a valid fault signal. prop Less than threshold T th When this is detected as transient interference, the current signal is filtered out. th Take a value of 0.3 to 0.

5.

8. The fault detection and handling method based on a floor cleaning robot according to claim 3, characterized in that, For each subsystem, the health parameters of that subsystem are calculated using the following formula: , of which S health : The current subsystem's health parameter, ranging from (0, 1]; ε: Health parameter decay coefficient, used to adjust the failure impact rate; W i : The severity weight of the i-th type of fault, pre-configured, with a value range of 0.2 to 1.0; N i T represents the cumulative number of times the i-th type of fault is triggered. i Fault duration factor; N max : The maximum permissible number of faults in the current subsystem; K s System stability coefficient; K t Time decay coefficient, used to reduce the weight of historical faults; Health parameter range: S health ∈[0.8, 1]: Normal; S health ∈[0.2, 0.8): Sub-health; S health ∈[0, 0.2): Functionality disabled; The methods for determining the fault type include: when the S of the power supply subsystem health < 0.2, or, the S of the navigation subsystem health <0.2 and the execution subsystem's S health If the value is less than 0.2, it is determined to be fault type 1, indicating complete failure. When the navigation subsystem S health < 0.2 and the S of the power supply subsystem health ≥ 0.2 and the S of the execution subsystem health If the value is ≥0.2, it is determined to be fault type 2, and navigation fails; When executing subsystem S health < 0.2 and the S of the power supply subsystem health ≥ 0.2 and the S of the navigation subsystem health If the value is ≥0.2, it is determined to be fault type 3, execution failure; Fault type locking: When a series of faults reach a threshold within a short period of time, the fault type is locked to prevent frequent switching; it is unlocked after a period of continuous normal operation or after a power-on self-test.

9. The fault detection and handling method based on a floor cleaning robot according to claim 3, characterized in that, The processing unit executes a corresponding differentiated processing strategy based on the fault type output by the fault type determination unit; specifically as follows: according to S health If the condition is determined to be sub-healthy, then an early warning, reporting, and continued operation will be implemented. If the fault type is determined to be 3, the cleaning mechanism will stop and automatic retraction will be triggered. If the fault is determined to be type 2, then automatic mode will be turned off and manual mode will be turned on. If the fault is determined to be type 1, then immediately activate emergency stop, power-off protection, and report a fatal fault.

10. A fault detection and handling system based on a floor cleaning robot, characterized in that, The system includes: a sensing and acquisition unit, a fault detection unit, a health parameter calculation unit, a fault type determination unit, a processing unit, and an interaction unit; The sensing and acquisition unit includes: lidar, ToF, binocular inertial module, IMU, power controller, temperature sensor, and liquid level sensor; the power controller is used to acquire current and voltage in real time. The fault detection unit includes: an energy characteristic detection module, a prediction residual detection module, a correlation characteristic detection module, and an inter-domain consistency detection module; The health parameter calculation unit includes: a navigation subsystem health parameter calculation module, an execution subsystem health parameter calculation module, and a power supply subsystem health parameter calculation module; Fault type determination unit: Based on the health parameters output by the health parameter calculation unit, output fault type 1, fault type 2, and fault type 3; Processing Unit: Executes the corresponding processing strategy based on the fault type output by the fault type determination unit; The interaction unit includes: navigation board, VCU, power board, BMS, cloud server, and state machine.

Images