A multi-level self-healing method and system for industrial cellular networks

Abstract

This invention discloses a multi-level self-healing method and system for industrial cellular networks, aiming to establish an interpretable and highly deterministic fault root cause diagnosis mechanism. It achieves accurate fault classification through fault fingerprinting based on a weight matrix, provides a multi-level self-healing strategy, and selects the optimal recovery scheme according to the fault type and cost. The method includes the following steps: deploying monitoring points at different levels, extracting the original indicators of each level, and discretizing the original indicators into feature state values; performing fault fingerprinting based on a weight matrix; and executing the corresponding self-healing action according to the fault fingerprint type. This invention belongs to the field of Internet of Things (IoT) technology.

Description

Technical Field

[0001] This invention belongs to the field of Internet of Things (IoT) technology, and more specifically, relates to a multi-level self-healing method and system for industrial cellular networks. Background Technology

[0002] In industrial IoT scenarios, cellular modules are often deployed in unattended environments. However, if a cellular network malfunctions, troubleshooting needs to be done promptly.

[0003] Existing technologies mostly rely on application-layer Ping to detect network connectivity issues. This approach has drawbacks, as it cannot distinguish between carrier network speed limits, APN configuration errors, and hardware failures. It often employs a one-size-fits-all restart strategy, which not only results in long recovery times but may also cause unnecessary service interruptions. Summary of the Invention

[0004] The main objective of this invention is to provide a multi-level self-healing method and system for industrial cellular networks, aiming to establish an interpretable and highly deterministic fault root cause diagnosis mechanism, achieve accurate fault classification through fault fingerprinting based on weight matrix, provide a multi-level self-healing strategy, and select the optimal recovery scheme according to the fault type and cost.

[0005] According to a first aspect of the present invention, a multi-level self-healing method for industrial cellular networks is provided, comprising the following steps:

[0006] Step 1: Deploy monitoring points at different levels, extract the original indicators for each level, and discretize the original indicators into feature state values. ; , =0 indicates normal. =1 indicates a warning. =2 indicates an anomaly;

[0007] The hierarchy includes the kernel driver layer. Bus Interaction Layer Protocol stack layer and module event layer ;

[0008] Step 2: Fault fingerprinting based on weight matrix:

[0009] Define fault fingerprint scoring Through characteristic state values With preset weights Weighted summation calculation:

[0010] ;

[0011] ;

[0012] ;

[0013] when At the same time, it was detected and and ;

[0014] when At the same time, it was detected

[0015] when At the same time, it was detected or and It is identified as fingerprint C;

[0016] when At the same time, it was detected and and It is identified as fingerprint D;

[0017] Step 3: Based on the type of faulty fingerprint, execute the corresponding self-healing action;

[0018] If it is fingerprint A, execute the first level of self-healing action, which is protocol layer self-healing.

[0019] If it is fingerprint B, execute the second level of self-healing action, which is logical link repair.

[0020] If it is fingerprint C, execute the third level of self-healing action, which is a hardware link switch.

[0021] If it is fingerprint D, execute the fourth level of self-healing action, which is a low-level physical reset.

[0022] The first to fourth levels of self-healing actions precisely mitigate the impact of faults, escalating gradually from low to high cost. The first level targets the OS network space, the second level targets the operator's core network connection, the third level targets the air interface radio frequency environment, and the fourth level targets the hardware entity / bus. The first and second levels of self-healing actions are logical resets that do not involve underlying hardware power supply oscillations. The third and fourth levels are hardware resets that cause current fluctuations and kernel driver overloads. Therefore, if a problem can be solved by a lower-level self-healing action, a higher-level self-healing action should not be used. This invention can select the optimal self-healing solution based on the fault fingerprint type, avoiding blind restarts and reducing service interruption time and system power consumption.

[0023] In the above-mentioned multi-level self-healing method for industrial cellular networks, for the kernel driver layer, the monitoring program is mounted on the dev_queue_xmit function through eBPF technology to monitor the network card driver's sending queue in real time.

[0024] If the queue backlog depth exceeds 80% and there is no consumption for 5 seconds... =2;

[0025] If the queue backlog depth is between 50% and 80%, or if eBPF detects a jitter of more than 3 times in skb dwell time, =1;

[0026] If the queue backlog depth is less than 50% =0.

[0027] In the above-mentioned multi-level self-healing method for industrial cellular networks, for the bus interaction layer, the data flow of the USB bus BulkIn / Out endpoints is monitored;

[0028] If the kernel Carrier state is 1 but there is no data exchange on the bus, and the bus throughput is 0 during the observation period, =2;

[0029] If the kernel Carrier state is 1, but the USB bus response latency exceeds 500ms for three consecutive times, =1;

[0030] If the USB endpoint data stream throughput meets business expectations. =0.

[0031] In the above-mentioned multi-level self-healing method for industrial cellular networks, for the protocol stack layer, the TCP retransmission count is obtained through netlink;

[0032] If the retransmission rate is between 15% and 30%, or if a continuous shrinking congestion window is detected, =1;

[0033] If the retransmission rate is >30% =2;

[0034] If the retransmission rate is less than 15% =0.

[0035] In the above-mentioned multi-level self-healing method for industrial cellular networks, for the module event layer, asynchronous URC events are monitored;

[0036] If an attach event, logout event, or SIM card removal event is captured... =2;

[0037] If a URC (Urban Reception Notice) indicating a drastic change in signal quality is detected, or a cell handover notice is detected, =1;

[0038] If the module reports +CREG:1 or +CEREG:1 =0.

[0039] In the aforementioned multi-level self-healing method for industrial cellular networks, For the weights of the kernel driver layer, =5;

[0040] For the weights of the bus interaction layer, =5;

[0041] For the weights of the protocol stack layer, =3;

[0042] For the weights of the module event layer, =8.

[0043] In the above-mentioned multi-level self-healing method for industrial cellular networks, the first-level self-healing action includes cleaning the kernel routing table and restarting the DHCP client;

[0044] The second level of self-healing action includes issuing AT+CGACT commands to re-establish the mobility management context;

[0045] The third level of self-healing action includes switching to a backup SIM card using backup base station information obtained through prior background scanning;

[0046] The fourth level of self-healing includes resetting the bus via the kernel's usb_reset_device; if that fails, it performs a hard reset of the module by operating the GPIO pins.

[0047] According to a second aspect of the present invention, an industrial cellular network multi-level self-healing system for implementing the method described in the first aspect is provided, comprising a feature extraction unit, a fault decision engine, and a self-healing actuator;

[0048] Feature extraction unit: Monitoring points are deployed at different levels to extract raw indicators from each level, and these raw indicators are discretized into feature state values. ; , =0 indicates normal. =1 indicates a warning. =2 indicates an anomaly;

[0049] The hierarchy includes the kernel driver layer. Bus Interaction Layer Protocol stack layer and module event layer ;

[0050] Fault Decision Engine: Define Fault Fingerprint Scoring Through characteristic state values With preset weights Weighted summation calculation:

[0051] ;

[0052] ;

[0053] ;

[0054] when At the same time, it was detected and and ;

[0055] when At the same time, it was detected

[0056] when At the same time, it was detected or and It is identified as fingerprint C;

[0057] when At the same time, it was detected and and It is identified as fingerprint D;

[0058] Self-healing actuator: Executes the corresponding self-healing action based on the type of fault fingerprint;

[0059] If it is fingerprint A, execute the first level of self-healing action, which is protocol layer self-healing.

[0060] If it is fingerprint B, execute the second level of self-healing action, which is logical link repair.

[0061] If it is fingerprint C, execute the third level of self-healing action, which is a hardware link switch.

[0062] If it is fingerprint D, execute the fourth level of self-healing action, which is a low-level physical reset.

[0063] In the aforementioned multi-level self-healing system for industrial cellular networks, the feature extraction unit includes an eBPF network probe, a USB / TTY driver monitor, a TCP / IP protocol stack counter, and a URC event listener.

[0064] eBPF network probes are used to monitor the network interface card (NIC) driver's transmit queue. If the queue backlog depth exceeds 80% and there is no data transmission for 5 seconds, the probe will detect the backlog. =2;

[0065] If the queue backlog depth is between 50% and 80%, or if eBPF detects a jitter of more than 3 times in skb dwell time, =1;

[0066] If the queue backlog depth is less than 50% =0;

[0067] The USB / TTY driver monitor is used to monitor the data flow at the USB bus Bulk In / Out endpoints. If the kernel Carrier state is 1 but there is no data exchange on the bus, and the bus throughput is 0 during the observation period, =2;

[0068] If the kernel Carrier state is 1, but the USB bus response latency exceeds 500ms for three consecutive times, =1;

[0069] If the USB endpoint data stream throughput meets business expectations. =0;

[0070] The TCP / IP protocol stack counter is used to obtain the TCP retransmission count. If the retransmission rate is between 15% and 30%, or if the congestion window is detected to be continuously shrinking, =1;

[0071] If the retransmission rate is >30% =2;

[0072] If the retransmission rate is less than 15% =0

[0073] The URC event listener is used to listen for asynchronous URC events. If an attach, unattach, or SIM card removal event is captured, it will be displayed. =2;

[0074] If a URC (Urban Reception Notice) indicating a drastic change in signal quality is detected, or a cell handover notice is detected, =1;

[0075] If the module reports +CREG:1 or +CEREG:1 =0.

[0076] In the aforementioned multi-level self-healing system for industrial cellular networks, For the weights of the kernel driver layer, =5;

[0077] For the weights of the bus interaction layer, =5;

[0078] For the weights of the protocol stack layer, =3;

[0079] For the weights of the module event layer, =8.

[0080] One of the above-described technical solutions of the present invention has at least one of the following advantages or beneficial effects:

[0081] In this invention, monitoring points are deployed at different levels, the extracted original indicators are discretized into feature state values, root cause judgment is made by fault fingerprint scoring based on weight matrix, and multi-level self-healing actions are executed according to the principle of cost minimization based on the judgment results.

[0082] It utilizes eBPF to monitor driver queues, solving the industry pain point that application layer Ping cannot distinguish between "system packet loss" and "network packet loss", and can identify "silent packet loss" and "driver freeze" that the application layer cannot perceive.

[0083] Based on the URC asynchronous listening mechanism, it ensures that self-healing can be initiated the instant network-side detachment occurs, which is 5 to 10 seconds faster than application layer detection; by monitoring module URC events, it can detect events such as network detachment and abnormal registration status in advance.

[0084] When SIM card switching is required, the backup base station information obtained by prior background scanning is used to skip the network search process and achieve SIM card switching in seconds.

[0085] It can select the optimal self-healing solution based on the fault fingerprint type, avoiding blind restarts and reducing business interruption time and system energy consumption. Attached Figure Description

[0086] The present invention will be further described below with reference to the accompanying drawings and embodiments;

[0087] Figure 1 This is a flowchart of the multi-level self-healing method for industrial cellular networks of the present invention;

[0088] Figure 2 This is a schematic diagram of the multi-level self-healing system for industrial cellular networks of the present invention. Detailed Implementation

[0089] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0090] This application deploys monitoring points at different levels, discretizes the extracted raw indicators into feature state values, performs root cause judgment through fault fingerprint scoring based on weight matrix, and executes multi-level self-healing actions according to the principle of cost minimization based on the judgment results.

[0091] Reference Figure 1As shown, a multi-level self-healing method for industrial cellular networks includes the following steps:

[0092] Step 1: Deploy monitoring points at different levels, extract the original indicators for each level, and discretize the original indicators into feature state values. , , =0 indicates normal. =1 indicates a warning. =2 indicates an anomaly;

[0093] The hierarchy includes the kernel driver layer. Bus Interaction Layer Protocol stack layer and module event layer ;

[0094] For monitoring the kernel driver layer, eBPF technology is primarily used to mount the monitoring program onto the `dev_queue_xmit` function, thereby enabling real-time monitoring of the network card driver's send queue. This method effectively monitors the packet backlog in the queue and judges the network transmission status and performance based on the backlog depth. If the queue backlog depth exceeds 80% and there is no data consumption for 5 seconds, the driver is determined to have entered a deadlock or apparent dead state. =2; If the queue backlog depth is between 50% and 80%, or if eBPF detects a jitter of more than 3 times in skb dwell time, the kernel driver layer processing speed slows down, but it is not completely stuck. =1; if the queue backlog depth is less than 50% and the processing latency is stable, then =0;

[0095] Monitoring of the bus interaction layer involves monitoring the data flow at the USB bus Bulk In / Out endpoints. If the kernel Carrier state is 1 but there is no data exchange on the bus, and the bus throughput is 0 during the observation period, it is determined that the physical bus is logically dead. =2; If the kernel Carrier state is 1, but the USB bus response latency (URB Completion Time) exceeds 500ms for three consecutive times, it is determined that there is communication interference in the bus protocol stack or the module response is sluggish. =1; If the USB endpoint data stream throughput meets business expectations. =0;

[0096] For monitoring the protocol stack layer, TCP retransmission counts are obtained via netlink. Based on these retransmission counts and the total transmission count, the retransmission rate can be calculated in real time. If the retransmission rate is between 15% and 30%, or if a continuous shrinking congestion window (CWND) is detected, a warning is issued. =1; If the retransmission rate is >30%, it is judged as extremely poor network link quality or APN link failure, and marked as abnormal. =2; if the retransmission rate is <15%, it is normal. =0;

[0097] Monitoring of the module's event layer is achieved by listening to the URC asynchronous events of " / dev / ttyUSB". If an unattach (+QIND: "pdpdeact") event, a logout (+CEREG: 0) event, or a SIM card removal event (+CPIN: NOTINSERTED) event is captured, it is determined that the module has completely disconnected from the network or there is a hardware malfunction. =2; If a sudden change in signal quality (URC) is detected (e.g., +CSQ drops sharply from a high score to a critical value), or a handover notification is detected, it is determined that the link is about to become unstable but has not yet been disconnected. =1; If the module reports +CREG:1 (registered home location) or +CEREG:1, the signal quality (RSRP) is stable. =0.

[0098] Step 2: Fault fingerprinting based on weight matrix:

[0099] Define fault fingerprint scoring Through characteristic state values With preset weights Weighted summation calculation:

[0100] ;

[0101] ;

[0102] ;

[0103] so, ;

[0104] This refers to the weight of the kernel driver layer, which is a key indicator of deadlock in the underlying software. =5;

[0105] The weights of the bus interaction layer are the core criteria for determining link connectivity. =5;

[0106] The weights of the protocol stack layers reflect link quality and are subject to fluctuations, hence their relatively low weights. =3;

[0107] The weights for the module event layer are as follows: the module event layer is actively reported by the network side and has the highest confidence level. =8;

[0108] To improve the accuracy of the judgment, in addition to calculating the fault fingerprint score... It is also necessary to detect core features, and the combination of the two can pinpoint the specific type of fault fingerprint.

[0109] when Further testing is needed. , and ,if and and The fingerprint is identified as A (logical block).

[0110] when Further testing is needed. If detected The fingerprint was identified as B (dial interruption).

[0111] when Further testing is needed. , and If detected and ,or and The fingerprint is identified as C (signal blind zone).

[0112] when Further testing is needed. , and If detected and and The fingerprint was identified as D (module dead).

[0113] Step 3: Based on the type of faulty fingerprint, execute the corresponding self-healing action;

[0114] If it is fingerprint A, execute the first level of self-healing action. The first level of self-healing action is protocol layer self-healing, which resolves the network outage caused by OS network stack state machine abnormality by clearing the kernel routing table and restarting the DHCP client. Specifically, execute iproute flush table main to clear the kernel routing table, execute dhclient -r and then execute dhclient to restart the DHCP client. After that, wait about 10 seconds and then verify network connectivity.

[0115] If it is fingerprint B, the second level of self-healing action is performed. The second level of self-healing action is logical link repair. The mobility management context is re-established by issuing AT+CGACT command without restarting the hardware. Specifically, the PDP context is re-established by issuing AT+CGACT and other commands through / dev / ttyUSB without restarting the hardware. After waiting for about 30 seconds, the connectivity is verified.

[0116] If it is fingerprint C, the third level of self-healing action is executed. The third level of self-healing action is hardware link switching. It uses the backup base station information (PCI / RSRP) obtained by prior background scanning to skip the network search process and achieve card switching in seconds. Specifically, it reads the backup link profile (PCI / RSRP) maintained by prior background scanning, switches to the backup SIM card slot through GPIO or AT commands, and enters the network according to the profile parameters, skipping the network search.

[0117] If it is fingerprint D, execute the fourth level self-healing action, which is a low-level physical reset. Attempt to reset the bus through the kernel usb_reset_device. If it fails, perform a hard reset (PowerCycle) on the module by operating the GPIO pin. Wait for the module to be powered on again and then redial. Verification is completed in about 120 seconds.

[0118] The mapping table between fault fingerprint types and self-healing actions is shown in Table 1:

[0119] Table 1. Mapping Table of Fault Fingerprint Types and Self-Healing Actions

[0120] Fault fingerprint type Core features Self-healing action level Self-healing action description Fingerprint A and and Level 1 Protocol layer repair (reset routing / DNS) Fingerprint B Level 2 Logical link repair (reactivation of PDP) Fingerprint C and( or ) Level 3 Hardware link switching (SIM card switching) Fingerprint D and and Level 4 Low-level physical reset (GPIO reset module)

[0121] In this application, eBPF is used to monitor the driver queue, which solves the industry pain point that application layer Ping cannot distinguish between "system packet loss" and "network packet loss", and can identify "silent packet loss" and "driver freeze" that the application layer cannot perceive.

[0122] The URC asynchronous listening mechanism ensures that self-healing can be initiated the instant network-side detachment occurs, which is 5 to 10 seconds faster than application layer detection. By monitoring module URC events, events such as network detachment and abnormal registration status can be detected in advance.

[0123] By using weighted summation instead of complex mathematical models, this method can run in real time on any low-power embedded processor without the need for floating-point units, resulting in extremely low computational overhead.

[0124] Through the fault fingerprint matrix, maintenance personnel can clearly trace the cause of restarts, which facilitates fault backtracking in unattended environments; a direct correlation is established between "monitoring indicators - judgment rules - physical actions";

[0125] By selecting the optimal self-healing solution from level one to level four based on the fault fingerprint type, blind restarts are avoided, reducing business interruption time and system energy consumption.

[0126] In this embodiment, the prior background scanning mechanism controls inactive card slots to perform background scanning during the gap period of normal communication on the current link and maintains a backup link profile. Then, if the determination result is fingerprint C, the network is directly entered according to the profile parameters, avoiding minute-level service interruption caused by searching the network again after switching.

[0127] Reference Figure 2 According to a second aspect of the present invention, a multi-level self-healing system for industrial cellular networks is provided, comprising a feature extraction unit, a fault decision engine, and a self-healing actuator;

[0128] Feature extraction unit: Monitoring points are deployed at different levels to extract raw indicators from each level, and these raw indicators are discretized into feature state values. ; , =0 indicates normal. =1 indicates a warning. =2 indicates an anomaly;

[0129] The hierarchy includes the kernel driver layer. Bus Interaction Layer Protocol stack layer and module event layer ;

[0130] The feature extraction unit includes an eBPF network probe, a USB / TTY driver monitor, a TCP / IP protocol stack counter, and a URC event listener;

[0131] eBPF probes transmit SKB / TX Queue information to the fault decision engine via an asynchronous reporting mechanism; eBPF network probes monitor the network card driver's send queue. If the queue backlog depth exceeds 80% and there is no data consumption for 5 seconds, the driver is determined to be in a deadlock or apparent dead state. =2; If the queue backlog depth is between 50% and 80%, or if eBPF detects a jitter of more than 3 times in skb dwell time, the kernel driver layer processing speed slows down, but it is not completely stuck. =1; if the queue backlog depth is less than 50% and the processing latency is stable, then =0;

[0132] The USB / TTY driver monitor is used to monitor the data flow at the USB bus Bulk In / Out endpoints. If the kernel Carrier status is 1 but there is no data exchange on the bus, and the bus throughput is 0 during the observation period, it is determined that the physical bus is logically dead. =2; If the kernel Carrier state is 1, but the USB bus response latency (URB Completion Time) exceeds 500ms for three consecutive times, it is determined that there is communication interference in the bus protocol stack or the module response is sluggish. =1; If the USB endpoint data stream throughput meets business expectations. =0; The USB / TTY driver monitor transmits the status change (Carrier / URB) to the fault decision engine;

[0133] The TCP / IP protocol stack counter is used to obtain the TCP retransmission count. If the retransmission rate is between 15% and 30%, or if the congestion window (CWND) is detected to be continuously shrinking, it is marked as a warning. =1; If the retransmission rate is >30%, it is judged as extremely poor network link quality or APN link failure, and marked as abnormal. =2; if the retransmission rate is <15%, it is normal. =0; The TCP / IP protocol stack counter transmits statistics (Retrans / CWND) to the fault decision engine;

[0134] The URC event listener is used to monitor asynchronous URC events. If it captures a detach (+QIND: "pdpdeact") event, a logout (+CEREG: 0) event, or a SIM card removal event (+CPIN: NOT INSERTED), it determines that the module has completely disconnected from the network or there is a hardware malfunction. =2; If a sudden change in signal quality (URC) is detected (e.g., +CSQ drops sharply from a high score to a critical value), or a handover notification is detected, it is determined that the link is about to become unstable but has not yet been disconnected. =1; If the module reports +CREG:1 (registered home location) or +CEREG:1, the signal quality (RSRP) is stable. =0; The URC event listener parses the URC event stream of the module control port in real time through the serial port listening thread and transmits events such as network registration status, signal strength, and EPS attachment status to the fault decision engine.

[0135] Fault Decision Engine: Define Fault Fingerprint Scoring Through characteristic state values With preset weights Weighted summation calculation:

[0136] The system calculates the fault fingerprint score in each sampling period (default 1 second). ;

[0137] ;

[0138] ;

[0139] For the weights of the kernel driver layer, =5;

[0140] For the weights of the bus interaction layer, =5;

[0141] For the weights of the protocol stack layer, =3;

[0142] For the weights of the module event layer, =8;

[0143] when At the same time, it was detected and and It is identified as fingerprint A;

[0144] when At the same time, it was detected Fingerprint B was identified.

[0145] when At the same time, it was detected ,or and It is identified as fingerprint C;

[0146] when At the same time, it was detected and and It is identified as fingerprint D;

[0147] Self-healing actuator: Executes the corresponding self-healing action based on the type of fault fingerprint;

[0148] If it is fingerprint A, execute the first level of self-healing action, which is protocol layer self-healing.

[0149] If it is fingerprint B, execute the second level of self-healing action, which is logical link repair.

[0150] If it is fingerprint C, execute the third level of self-healing action, which is a hardware link switch.

[0151] If it is fingerprint D, execute the fourth level of self-healing action, which is a low-level physical reset.

[0152] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A multi-level self-healing method for industrial cellular networks, characterized in that, Includes the following steps: Step 1: Deploy monitoring points at different levels, extract the original indicators for each level, and discretize the original indicators into feature state values. ; , =0 indicates normal. =1 indicates a warning. =2 indicates an anomaly; The hierarchy includes the kernel driver layer. Bus Interaction Layer Protocol stack layer and module event layer ; Step 2: Fault fingerprinting based on weight matrix: Define fault fingerprint scoring Through characteristic state values With preset weights Weighted summation calculation: ； ； ； when At the same time, it was detected and and It is identified as fingerprint A; when At the same time, it was detected Fingerprint B was identified. when At the same time, it was detected ,or and It is identified as fingerprint C; when At the same time, it was detected and and It is identified as fingerprint D; Step 3: Based on the type of faulty fingerprint, execute the corresponding self-healing action; If it is fingerprint A, execute the first level of self-healing action, which is protocol layer self-healing. If it is fingerprint B, execute the second level of self-healing action, which is logical link repair. If it is fingerprint C, execute the third level of self-healing action, which is a hardware link switch. If it is fingerprint D, execute the fourth level of self-healing action, which is a low-level physical reset.

2. The multi-level self-healing method for industrial cellular networks according to claim 1, characterized in that, For the kernel driver layer, the monitoring program is mounted on the dev_queue_xmit function using eBPF technology to monitor the network card driver's send queue in real time. If the queue backlog depth exceeds 80% and there is no consumption for 5 seconds... =2; If the queue backlog depth is between 50% and 80%, or if eBPF detects a jitter of more than 3 times in skb dwell time, =1; If the queue backlog depth is less than 50% =0.

3. The multi-level self-healing method for industrial cellular networks according to claim 1, characterized in that, For the bus interaction layer, monitor the data flow at the USB bus Bulk In / Out endpoints; If the kernel Carrier state is 1 but there is no data exchange on the bus, and the bus throughput is 0 during the observation period, =2; If the kernel Carrier state is 1, but the USB bus response latency exceeds 500ms for three consecutive times, =1; If the USB endpoint data stream throughput meets business expectations. =0.

4. The multi-level self-healing method for industrial cellular networks according to claim 1, characterized in that, For the protocol stack layer, the TCP retransmission count is obtained through netlink; If the retransmission rate is between 15% and 30%, or if a continuous shrinking congestion window is detected, =1; If the retransmission rate is >30% =2; If the retransmission rate is less than 15% =0.

5. The multi-level self-healing method for industrial cellular networks according to claim 1, characterized in that, For the module event layer, asynchronous URC events are monitored; If an attach event, logout event, or SIM card removal event is captured... =2; If a URC (Urban Reception Notice) indicating a drastic change in signal quality is detected, or a cell handover notice is detected, =1; If the module reports +CREG:1 or +CEREG:1 =0.

6. The multi-level self-healing method for industrial cellular networks according to claim 1, characterized in that, For the weights of the kernel driver layer, =5; For the weights of the bus interaction layer, =5; For the weights of the protocol stack layer, =3; For the weights of the module event layer, =8.

7. The multi-level self-healing method for industrial cellular networks according to claim 1, characterized in that, The first level of self-healing actions includes cleaning the kernel routing table and restarting the DHCP client; The second level of self-healing action includes issuing AT+CGACT commands to re-establish the mobility management context; The third level of self-healing action includes switching to a backup SIM card using backup base station information obtained through prior background scanning; The fourth level of self-healing includes resetting the bus via the kernel's usb_reset_device; if that fails, it performs a hard reset of the module by operating the GPIO pins.

8. A multi-level self-healing system for industrial cellular networks used to implement the method of any one of claims 1-7, characterized in that, It includes a feature extraction unit, a fault decision engine, and a self-healing executor; Feature extraction unit: Monitoring points are deployed at different levels to extract raw indicators from each level, and these raw indicators are discretized into feature state values. ; , =0 indicates normal. =1 indicates a warning. =2 indicates an anomaly; The hierarchy includes the kernel driver layer. Bus Interaction Layer Protocol stack layer and module event layer ; Fault Decision Engine: Define Fault Fingerprint Scoring Through characteristic state values With preset weights Weighted summation calculation: ； ； ； when At the same time, it was detected and and ; when At the same time, it was detected when At the same time, it was detected or and It is identified as fingerprint C; when At the same time, it was detected and and It is identified as fingerprint D; Self-healing actuator: Executes the corresponding self-healing action based on the type of fault fingerprint; If it is fingerprint A, execute the first level of self-healing action, which is protocol layer self-healing. If it is fingerprint B, execute the second level of self-healing action, which is logical link repair. If it is fingerprint C, execute the third level of self-healing action, which is a hardware link switch. If it is fingerprint D, execute the fourth level of self-healing action, which is a low-level physical reset.

9. The industrial cellular network multi-level self-healing system according to claim 8, characterized in that, The feature extraction unit includes an eBPF network probe, a USB / TTY driver monitor, a TCP / IP protocol stack counter, and a URC event listener; eBPF network probes are used to monitor the network interface card (NIC) driver's transmit queue. If the queue backlog depth exceeds 80% and there is no data transmission for 5 seconds, the probe will detect the backlog. =2; If the queue backlog depth is between 50% and 80%, or if eBPF detects a jitter of more than 3 times in skb dwell time, =1; If the queue backlog depth is less than 50% =0; The USB / TTY driver monitor is used to monitor the data flow at the USB bus Bulk In / Out endpoints. If the kernel Carrier state is 1 but there is no data exchange on the bus, and the bus throughput is 0 during the observation period, =2; If the kernel Carrier state is 1, but the USB bus response latency exceeds 500ms for three consecutive times, =1; If the USB endpoint data stream throughput meets business expectations. =0; The TCP / IP protocol stack counter is used to obtain the TCP retransmission count. If the retransmission rate is between 15% and 30%, or if the congestion window is detected to be continuously shrinking, =1; If the retransmission rate is >30% =2; If the retransmission rate is less than 15% =0; The URC event listener is used to listen for asynchronous URC events. If an attach, unattach, or SIM card removal event is captured, it will be displayed. =2; If a URC (Urban Reception Notice) indicating a drastic change in signal quality is detected, or a cell handover notice is detected, =1; If the module reports +CREG:1 or +CEREG:1 =0.

10. The industrial cellular network multi-level self-healing system according to claim 8, characterized in that, For the weights of the kernel driver layer, =5; For the weights of the bus interaction layer, =5; For the weights of the protocol stack layer, =3; For the weights of the module event layer, =8.

Images