An adaptive environment alarm and status indication system for a battery production plant

Abstract

The present application relates to the technical field of battery production environment monitoring and industrial safety alarm, in particular to a self-adaptive environment alarm and state indication system for battery production workshop, comprising a data acquisition module for acquiring multi-source environment state data with time stamp and regional position identification and historical alarm interaction data of the monitoring area; a pre-warning generation module for generating initial probability pre-warning signal based on the multi-source environment state data; a fatigue assessment module for calculating alarm fatigue index based on the trigger frequency of the initial probability pre-warning signal and the historical alarm interaction data; a self-adaptive indication module for outputting initial probability pre-warning signal when the alarm fatigue index is lower than the preset fatigue threshold, and generating global macro state indication signal when the alarm fatigue index is not lower than the preset fatigue threshold; an alarm execution module for controlling the display state of the state indication board; the present application realizes the closed-loop balance of environment pre-warning and personnel responsiveness.

Description

Technical Field

[0001] This invention relates to the field of battery production environment monitoring and industrial safety alarm technology, specifically an adaptive environmental alarm and status indication system for battery production workshops. Background Technology

[0002] As a manufacturing environment with high requirements for environmental stability, battery production workshops typically need to continuously acquire multiple environmental parameters in areas such as drying control, dust control, and volatile gas monitoring. These parameters are then used to provide early warnings to on-site personnel through status indicator dashboards. Therefore, the rationality of the alarm mechanism and status expression method directly affects the effectiveness of workshop safety management and personnel response efficiency.

[0003] Existing environmental alarm methods have many problems. For example, they often use fixed threshold judgment and a single strong prompt output mode. Although they can alert to abnormalities in dew point, dust, or volatile gases, in continuous production and high-frequency fluctuation scenarios, they are prone to repeated triggering of alarms in multiple areas and with multiple parameters. This leads to frequent flashing of dashboards, excessive information density, and problems such as delayed confirmation, increased silence, and even decreased sensitivity to alarm signals for operators. At the same time, existing solutions usually lack fatigue assessment and adaptive switching mechanisms based on historical interaction behavior, making it difficult to balance fine-grained trend warnings and macroscopic, easily understandable expressions. Furthermore, in cases of data loss, partial missing measurements, or physical confirmation of disasters, existing systems often lack stable compensation processing and forced blocking capabilities. Therefore, it is difficult to balance the accuracy of continuous warnings, the acceptability of long-term use, and the timeliness of handling extreme risks. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides an adaptive environmental alarm and status indication system for battery production workshops. Specifically, the technical solution of this invention includes:

[0005] The data acquisition module is used to acquire multi-source environmental status data with timestamps and regional location identifiers for the monitored area, as well as historical alarm interaction data;

[0006] The early warning generation module is used to generate an initial probability early warning signal based on the multi-source environmental state data and through a preset anomaly prediction model.

[0007] The fatigue assessment module is used to calculate the alarm fatigue index based on the triggering frequency of the initial probability warning signal within a preset time period and the historical alarm interaction data.

[0008] An adaptive indication module is used to determine the relationship between the alarm fatigue index and a preset fatigue threshold. If the alarm fatigue index is lower than the preset fatigue threshold, the initial probability warning signal is used as the target state indication signal. If the alarm fatigue index is not lower than the preset fatigue threshold, the initial probability warning signal is subjected to dimensionality reduction and reconstruction processing to generate a global macroscopic state indication signal, and the global macroscopic state indication signal is used as the target state indication signal.

[0009] The alarm execution module is used to send control commands to an external status indicator board based on the target status indication signal, so as to control the display status of the status indicator board.

[0010] Optionally, the early warning generation module includes: a noise filtering unit, used to filter the multi-source environmental state data based on a dynamic baseline drift algorithm that calculates the mean within a local time window and performs mean removal processing, to generate high signal-to-noise ratio environmental data; and a probability prediction unit, used to input the high signal-to-noise ratio environmental data as a deep reinforcement learning model of the preset anomaly prediction model, and output the initial probability early warning signal, wherein the initial probability early warning signal carries a state fluctuation probability value.

[0011] Optionally, the fatigue assessment module is specifically used to: extract the historical alarm confirmation delay time and historical mute operation frequency from the historical alarm interaction data; calculate the signal jump rate of the initial probability warning signal within the preset time period; normalize the historical alarm confirmation delay time, the historical mute operation frequency, and the signal jump rate; and perform a weighted summation of the normalized historical alarm confirmation delay time, the historical mute operation frequency, and the signal jump rate to calculate the alarm fatigue index.

[0012] Optionally, when the adaptive indication module performs dimensionality reduction and reconstruction processing on the initial probability warning signal, it specifically performs the following: masking the state fluctuation probability value in the initial probability warning signal; dividing the monitoring area into a preset number of sub-regions, and extracting target environmental parameters that exceed a preset safety hard threshold from the multi-source environmental state data within the sub-regions; determining the target environmental parameters as abnormal states, and performing Boolean logic OR operations on all target environmental parameters in abnormal states to generate the global macroscopic state indication signal.

[0013] Optionally, the alarm execution module includes: a first driving unit, configured to drive the status indicator panel to perform alternating high-frequency flashing display of multiple regions and multiple parameters in response to the target status indication signal being the initial probability warning signal; and a second driving unit, configured to drive the status indicator panel to perform uniform constant-on display of the entire region in response to the target status indication signal being the global macro status indication signal, and to shield the display of fluctuation values ​​of underlying parameters.

[0014] Optionally, the system further includes a disaster blocking module, which is used to: acquire a disaster physical confirmation signal of the monitoring area through an external fire linkage interface; determine whether the disaster physical confirmation signal contains a preset highest-level disaster identifier; if the disaster physical confirmation signal contains the highest-level disaster identifier, bypass the adaptive indication module, generate a highest-level evacuation command, and forcibly trigger an external global audible and visual alarm device; if the disaster physical confirmation signal does not contain the highest-level disaster identifier, maintain the current operating state of the alarm execution module.

[0015] Optionally, the data acquisition module further includes a data compensation unit, which is used to: monitor the data packet loss rate during the acquisition of the multi-source environmental state data; determine the relationship between the data packet loss rate and a preset packet loss threshold and a preset distortion upper limit; if the data packet loss rate is higher than the preset packet loss threshold but not higher than the preset distortion upper limit, then a time-series interpolation algorithm based on linear calculation of adjacent normal data points is used to reconstruct the missing data in the multi-source environmental state data, and the reconstructed data is input as the multi-source environmental state data into the early warning generation module; if the data packet loss rate is higher than the preset distortion upper limit, then forward hold or first-order gradient extrapolation interpolation is used to reconstruct the missing data in the multi-source environmental state data, and the reconstructed data is input as the multi-source environmental state data into the early warning generation module.

[0016] Optionally, the system further includes a threshold update module, which is used to: obtain missed detection feedback records from external operator terminals, and calculate the actual missed detection rate of the status indicator dashboard during the period when it outputs the global macroscopic status indicator signal based on the missed detection feedback records; determine the relationship between the actual missed detection rate and a preset missed detection tolerance; if the actual missed detection rate is higher than the preset missed detection tolerance, then lower the preset fatigue threshold; if the actual missed detection rate is not higher than the preset missed detection tolerance, then maintain the preset fatigue threshold.

[0017] Optionally, the monitoring area is a battery production workshop; the multi-source environmental status data includes cross-environmental parameters collected by dew point sensors, dust sensors, and volatile gas sensors distributed in the battery production workshop; and the historical alarm interaction data comes from the feedback operation records of the battery production workshop operators on the status indicator dashboard.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] 1. This system calculates the alarm fatigue index based on the trigger frequency of the initial probability warning signal and historical alarm interaction data through a fatigue assessment module and an adaptive indication module. When the alarm fatigue index is lower than the preset fatigue threshold, the status indicator dashboard uses multi-area, multi-parameter alternating high-frequency flashing to provide fine-grained early warning. When the index is not lower than the threshold, the system automatically performs dimensionality reduction and reconstruction processing on the signal, such as shielding the probability value of status fluctuations, and switches to a global macroscopic status indicator signal that is uniformly lit throughout the entire area. This mechanism effectively solves the problems of delayed operator confirmation, desensitization, and fatigue caused by frequent flashing and excessive information density in existing environmental alarm systems, and maintains a long-term usable human-machine interaction response while ensuring the early warning function.

[0020] 2. This system incorporates a data compensation unit and a noise filtering unit at the front end. When the packet loss rate of multi-source environmental state data acquisition exceeds the preset packet loss threshold, a temporal interpolation algorithm is used to reconstruct the missing data. Simultaneously, a dynamic baseline drift algorithm is employed for filtering, extracting high signal-to-noise ratio environmental data before inputting it into the deep reinforcement learning model. Furthermore, the system also includes a threshold update module that dynamically lowers the preset fatigue threshold based on the actual false negative rate. This design overcomes the shortcomings of existing technologies that are prone to computational distortion or false alarms when facing local missing data, data packet loss, and slow baseline drift, ensuring the continuity of input data and the accuracy of model prediction, and achieving a dynamic balance between security and understandability.

[0021] 3. This system has been specially equipped with a disaster prevention module, which can obtain physical confirmation signals of disasters in the monitoring area through an external fire linkage interface. Once the signal is determined to contain a preset highest-level disaster indicator, the system will immediately bypass the dimensionality reduction reconstruction and fatigue judgment link of the adaptive indication module, directly generate the highest-level evacuation command and forcibly trigger the external full-area audible and visual alarm device. This mechanism effectively makes up for the monitoring gaps of the existing adaptive alarm control logic in such extreme situations, prevents delays in evacuation due to fatigue-based downgrade display strategies when real disasters occur, sets preset safety constraints for high-risk battery production workshops, and ensures timely response under extreme risks. Attached Figure Description

[0022] Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0024] like Figure 1 As shown, an adaptive environmental alarm and status indication system for a battery production workshop includes:

[0025] The data acquisition module is used to acquire multi-source environmental status data with timestamps and regional location identifiers for the monitored area, as well as historical alarm interaction data;

[0026] The early warning generation module is used to generate an initial probability early warning signal based on multi-source environmental status data and a preset anomaly prediction model.

[0027] The fatigue assessment module is used to calculate the alarm fatigue index based on the triggering frequency of the initial probability warning signal within a preset time period and historical alarm interaction data.

[0028] The adaptive indication module is used to determine the relationship between the alarm fatigue index and the preset fatigue threshold. If the alarm fatigue index is lower than the preset fatigue threshold, the initial probability warning signal is used as the target state indication signal. If the alarm fatigue index is not lower than the preset fatigue threshold, the initial probability warning signal is subjected to dimensionality reduction and reconstruction to generate a global macroscopic state indication signal, and the global macroscopic state indication signal is used as the target state indication signal.

[0029] The alarm execution module is used to send control commands to an external status indicator board based on the target status indication signal, so as to control the display status of the status indicator board.

[0030] This embodiment provides an adaptive environmental alarm and status indication mechanism for battery production workshops. Specifically, the system is deployed in a status indication dashboard network above the coating area, assembly area and liquid injection area of ​​the lithium battery production workshop. It completes the entire closed-loop process of environmental data acquisition, trend warning, fatigue assessment, status degradation and dashboard display control, focusing on the main line of continuous high-risk operation but avoiding the reduction of personnel's sensitivity to alarm response.

[0031] Specifically, the data acquisition module receives environmental status data from multiple sensor nodes, and simultaneously obtains historical alarm interaction data from the dashboard interaction terminal, mute button, confirmation button, and team workstation terminal; the multi-source environmental status data here may include at least dew point value, dust concentration, volatile gas concentration, and corresponding timestamp and area number.

[0032] For ease of explanation, assume that data from three sub-regions are collected at a certain moment: the first region has a dew point of 0.93%, dust concentration of 32, and VOC concentration of 11; the second region has a dew point of 0.98%, dust concentration of 30, and VOC concentration of 10; and the third region has a dew point of 0.96%, dust concentration of 36, and VOC concentration of 13. Simultaneously, historical alarm interaction data records the number of times operators acknowledged alarms, the number of times alarms were silenced, and the average response delay within the last 30 minutes. After receiving the above multi-source environmental state data, the early warning generation module calls a preset anomaly prediction model to generate an initial probability early warning signal. This signal is not a simple binary result of alarm / no alarm, but rather carries a probability value of state fluctuation.

[0033] For example, the first region generates 0.41, the second region generates 0.76, and the third region generates 0.68, representing the probability that the corresponding region will enter an abnormal state within the subsequent short time window; the short time window here can be set to any of the following 2 minutes, 5 minutes, or 10 minutes; the fatigue assessment module further calculates the alarm fatigue index based on the triggering frequency of the initial probability warning signal within the preset time period, combined with historical alarm interaction data.

[0034] For a specific example analysis, assume the preset time period is the past 20 minutes, during which 12 probability warnings occurred, including 5 high-level warnings exceeding 0.7; meanwhile, the average operator confirmation delay was 18 seconds, and there were 4 silent operations; the system can map these characteristics to a unified score space, for example, obtaining an alarm fatigue index of 0.72, while the preset fatigue threshold is 0.65; the adaptive indication module determines the subsequent dashboard expression strategy based on this comparison result; when the alarm fatigue index is lower than the preset fatigue threshold, it indicates that the personnel still have an acceptable response capability to complex warnings, at which point the initial probability warning signal is directly used as the target status indication signal, so that the dashboard retains fine-grained risk expression for each area and parameter;

[0035] When the alarm fatigue index is not lower than the preset fatigue threshold, it indicates that the frequent flashing and high-density changes have approached the upper limit of what personnel can tolerate. At this point, the high-frequency fine-grained status will no longer be output. Instead, the initial probability warning signal will be reconstructed by reducing its dimensions and transformed into a global macro-state indication signal, such as the entire workshop entering an attention state or the entire area entering an alert state. The alarm execution module sends control commands to the status indication dashboard based on the target status indication signal. If the target status indication signal comes from the fine-grained probability warning, the dashboard will be displayed according to the region and parameters. If the target status indication signal comes from the reduced macro-state, the dashboard will be displayed with a uniform color and a stable rhythm to suppress the visual burden caused by the repeated jumps of the underlying fluctuation values.

[0036] In an optional anomaly handling mechanism, two types of boundary situations may occur during actual operation. First, if the multi-source environmental status data is incomplete at the current moment but still meets the minimum computability condition, where the minimum computability condition means that the number of types of valid environmental parameters currently acquired reaches the minimum feature dimension threshold required for the forward inference of the early warning model, for example, at least two of the three key parameters are valid, then the system continues to generate an early warning signal with reduced credibility and internally marks the data as insufficient. Second, if historical alarm interaction data is missing, for example, if a newly commissioned shift has not yet formed enough interaction records, the fatigue assessment module can be started with the default initial value or the baseline parameters of the nearest neighboring shift to prevent the entire data processing link from being interrupted due to missing interaction data.

[0037] In the electrolyte filling workshop of a large battery manufacturing plant, continuous rainy weather caused an increase in external humidity load, and the workshop's environmental control system operated near the process limit parameters for a long time. Between 10:00 and 10:20 a.m., the second and third zones experienced frequent short-term fluctuations in dew point and VOC due to local airflow changes and the volatilization of the new electrolyte. The warning generation module continuously gave probability signals close to 0.7, which the operators could still check one by one in the first 10 minutes. However, as the high-frequency changes on the dashboard continued and silent behavior increased, the alarm fatigue index rose to 0.72. After that, the system automatically switched the dashboard from a fine-grained mode of flashing by zone and scrolling probability values ​​to a macro mode of constant yellow light throughout the workshop, prompting manual inspection, thereby preventing the work team from losing attention to the complex signals.

[0038] The purpose of this step is to introduce the constraint variable of human responsiveness into the safety early warning link, so that the system no longer simply pursues the number of alarms, but maintains a long-term usable human-machine response relationship while ensuring the bottom line of safety, thereby achieving highly reliable adaptive alarm control for high-risk continuous production scenarios.

[0039] In this embodiment, the early warning generation module includes: a noise filtering unit, used to filter multi-source environmental state data based on a dynamic baseline drift algorithm that calculates the mean within a local time window and performs mean removal processing, to generate high signal-to-noise ratio environmental data; and a probability prediction unit, used to input the high signal-to-noise ratio environmental data as a deep reinforcement learning model of a preset anomaly prediction model, and output an initial probability early warning signal, wherein the initial probability early warning signal carries a state fluctuation probability value.

[0040] This embodiment provides a noise filtering and probability prediction mechanism for early warning generation. Specifically, when the aforementioned workshop is in a critical environmental control state for a long time, if the original sensor data is directly judged as abnormal, the baseline drift is easily mistaken for risk deterioration, resulting in the dashboard being in a high-frequency warning state for a long time. Therefore, this embodiment adds dynamic baseline drift filtering before generating early warnings and then performs probability prediction.

[0041] Specifically, the noise filtering unit uses the local time window of each sensor node as the calculation unit. For example, using the most recent 5 sampling points as a window, a dew point sensor sequentially collects 0.92, 0.93, 0.94, 0.95, and 1.00; the window mean is 0.948. After the system performs mean removal processing on the current window, it obtains the relative fluctuation sequence -0.028, -0.018, -0.008, 0.002, and 0.052. If 1.01 is subsequently collected, the window slides forward to 0.93, 0.94, 0.95, 1.00, and 1.01, and the mean changes to 0.966.

[0042] By using this time-sliding mean-removal processing, the system can distinguish between the overall rise of the sensor and short-term anomalous sudden changes; the former is more characterized by a slow increase in the mean, while the latter is characterized by a sudden increase after mean removal. The same approach can be used for dust and VOC sensors; assuming a VOC node has values ​​of 10, 10, 11, 11, and 14 at 5 sampling points, with a mean of 11.2, after mean removal, we get -1.2, -1.2, -0.2, -0.2, and 2.8; at this point, the deviation value of 2.8 highlights the short-term anomaly more effectively and is not masked by the preceding background value. The resulting high signal-to-noise ratio environmental data retains the risk trend while reducing the slow baseline drift caused by factors such as seasonal moisture infiltration and dust adhesion; the probability prediction unit inputs this high signal-to-noise ratio environmental data as a deep reinforcement learning model for the preset anomaly prediction model;

[0043] It is important to clarify that the online inference process and the model training process are decoupled in this part of the system. During the offline training or asynchronous update phase, the model takes the most recent window features of each region as the state input, outputs different levels of state fluctuation probability values ​​as actions, and maps the subsequent actual anomaly triggering and false alarm situations together as reward signals. The calculation logic of mapping to reward signals specifically includes: assigning positive reward values ​​to actual anomaly hits, assigning negative penalty values ​​to false alarms, and dynamically weighting the positive reward values ​​according to the time advance between the timestamp of the model output action and the timestamp of the subsequent actual anomaly occurrence.

[0044] Without interfering with the subsequent adaptive indication logic, the policy network parameters are iteratively optimized. In the real-time prediction stage executed by the current module, the probability prediction unit only calls the converged policy network for forward propagation calculation and no longer waits for feedback signals in real time. It directly outputs the probability distribution corresponding to each action based on the high signal-to-noise ratio characteristic state of the input, and directly maps the distribution to the state fluctuation probability value.

[0045] For ease of understanding, assume that during online operation, the model receives processed feature vectors from the first, second, and third regions. The first region exhibits low fluctuations, the second region shows a simultaneous increase in dew point and VOC, and the third region experiences a dust fluctuation exceeding the preset normal range. The output state fluctuation probability values ​​are 0.35, 0.79, and 0.62, respectively, thus forming an initial probability warning signal. In an optional anomaly handling mechanism, if the number of effective sampling points within a local time window is insufficient (e.g., the sensor has just come online or only two points are received due to communication jitter), the system can use a shortened window approach for transitional calculations, such as temporarily reducing the 5-point window to a 3-point window. If even the minimum window is insufficient, the node outputs a low-confidence flag, which is then replaced with a missing measurement filler value in subsequent model inputs to prevent the entire dashboard from entering a distorted state due to a single point of failure.

[0046] Furthermore, if a channel exhibits consistently zero variance or extremely low variance after mean removal, it can be determined that the sensor may be experiencing a sampling failure or jamming. In this case, a separate report of abnormal bitstream health should be submitted instead of mistaking it for an absolutely stable environment.

[0047] In the second shift after the aforementioned injection workshop entered a continuously high humidity environment, the overall dew point value near the outer wall rose to around 0.96 at a gradient lower than the preset rate of change. If the original values ​​were directly input into the prediction model, it would easily cause multiple areas to trigger high-probability alarms simultaneously. However, after the mean was removed, this slow upward movement was regarded as a background change. What truly triggered the high risk was a local area that abruptly changed from a relative deviation of 0.01 to 0.06 within two minutes, accompanied by a synchronous increase in VOC deviation. Therefore, the model only rated the second area as having a high fluctuation probability of 0.79, rather than classifying the entire workshop as abnormal.

[0048] The purpose of this mechanism is to first eliminate false abnormal signals in high-noise environments, and then use a time-series decision model to output probabilistic early warning results that are more consistent with the evolution of the scene, thereby achieving front-end interference suppression in high-density signal concurrent scenarios.

[0049] In this embodiment, the fatigue assessment module is specifically used to: extract historical alarm confirmation delay time and historical mute operation frequency from historical alarm interaction data; calculate the signal jump rate of the initial probability warning signal within a preset time period; normalize the historical alarm confirmation delay time, historical mute operation frequency, and signal jump rate; and calculate the alarm fatigue index by weighting and summing the normalized historical alarm confirmation delay time, historical mute operation frequency, and signal jump rate.

[0050] This embodiment provides a quantitative mechanism for fatigue assessment. Specifically, measuring staff workload solely by the number of warnings is insufficient, because with the same number of warnings, if timely confirmation and no one silences the warnings, it indicates that the work group is still under control; conversely, if confirmation is delayed and frequent silences occur, it indicates that the way the signboard is presented has caused cognitive overload on the staff. Therefore, this embodiment further introduces three quantitative factors: confirmation delay, silence frequency, and signal jump rate.

[0051] Specifically, the system extracts the alarm confirmation delay time from historical alarm interaction data; this delay time can be defined as the time interval between a warning issued by the dashboard and the nearest on-duty operator performing a confirmation action; assuming there are 4 events requiring confirmation in the past 20 minutes, with delays of 8s, 12s, 20s and 24s respectively, the average confirmation delay is 16s; the system also extracts the frequency of silent operations within the same period, assuming a total of 3 occurrences; at the same time, the fatigue assessment module calculates the signal transition rate of the initial probability warning signal;

[0052] To illustrate this more clearly, assume that the second area outputs a status level every 1 minute within 10 minutes, categorized into low, medium, and high based on probability values. The results would be: medium, high, medium, high, high, medium, low, medium, high, medium. There are a total of 8 level changes between adjacent samples. If we use the total number of comparisons (9) as the denominator, the jump rate is 8 / 9. The higher the jump rate, the more unstable the information on the dashboard, and the more difficult it is for the operator to form a continuous understanding.

[0053] In calculation, these three indicators can be normalized to a unified dimension first, and then summed according to preset weights; for example, the confirmation delay is normalized to 0.4, the silence frequency is normalized to 0.6, and the signal jump rate is normalized to 0.89; assuming the corresponding weights are 0.3, 0.2, and 0.5 respectively, the alarm fatigue index is... If the preset fatigue threshold is set to 0.65, the system determines that the current shift has entered a high-load state. Weighted summation is used here instead of single-index judgment because the sensitive factors are different for different workshops and shifts. For example, a response speed lower than the preset standard indicates that the work is more busy. On the other hand, some shifts have low confirmation delays but frequent silences, which indicates that there is a blocking behavior for high-frequency prompts that is higher than the preset frequency. Through weight adjustment, the system can adapt to different production organization forms.

[0054] In abnormal situations, if no confirmation action is recorded within a certain period, the confirmation delay cannot be directly recorded as zero, otherwise it will mask the risk of no response. In this case, it can be recorded as a preset timeout value or marked as an unconfirmed event. If the mute button is disabled due to maintenance, the mute frequency component can be temporarily set to invalid, and the weight of the remaining valid indicators can be normalized proportionally. If the denominator of the signal jump rate is too small, for example, only one sampling occurred, the statistical period should be extended before calculation to avoid amplifying accidental jitter.

[0055] During the afternoon hours in the aforementioned injection workshop, extreme weather caused slight fluctuations in multiple areas; the indicator colors on the dashboard switched between different warning levels every tens of seconds. Although the shift did not immediately stop the machine, response delays gradually appeared; system statistics showed that the average confirmation delay in the past 20 minutes increased from 7 seconds in the morning shift to 16 seconds, the number of silent operations increased from 0 to 3, and the status levels of the second and third areas switched frequently, with a jump rate close to 0.89; after comprehensive calculation, the alarm fatigue index exceeded the preset fatigue threshold, which became the trigger for subsequent display downgrade;

[0056] The purpose of this mechanism is to transform the characteristic of personnel response load into a calculable index, thereby enabling the alarm system to quantify the on-site cognitive load in real time.

[0057] In this embodiment, when the adaptive indication module performs dimensionality reduction and reconstruction processing on the initial probability warning signal, it specifically performs the following: masking the state fluctuation probability value in the initial probability warning signal; dividing the monitoring area into a preset number of sub-regions and extracting target environmental parameters that exceed a preset safety hard threshold from the multi-source environmental state data within the sub-regions; determining the target environmental parameters as abnormal states and performing Boolean logic OR operations on all target environmental parameters in abnormal states to generate a global macroscopic state indication signal.

[0058] This embodiment provides an information dimensionality reduction and reconstruction mechanism for high-load scenarios. Specifically, although the aforementioned probabilistic early warning can provide earlier trend recognition, when the dashboard changes frequently over a long period of time, operators are more likely to be disturbed by continuously changing probability values ​​and find it difficult to grasp whether an unacceptable safety boundary has been reached. Therefore, this embodiment actively masks the expression of probability values ​​after the fatigue index exceeds the limit and uses hard thresholds and Boolean synthesis to form a more intuitive global state.

[0059] Specifically, the first step in dimensionality reduction and reconstruction is to mask the state fluctuation probability values ​​in the initial probability warning signal; in other words, continuous values ​​such as 0.61, 0.73, and 0.68 are no longer output to the dashboard or drive changes in the flashing frequency of the zones. This is because, during high-load phases, small fluctuations in probability values ​​are not conducive to on-site understanding. The second step is to divide the monitoring area into several sub-areas according to the workshop layout; for example, it can be divided into four sub-areas: the first, second, third, and fourth, with each sub-area monitoring three parameters: dew point, dust, and VOCs. The system extracts target environmental parameters that exceed the safety hard threshold from the raw or pre-processed multi-source environmental data.

[0060] The hard thresholds here can be set by process safety requirements, such as a dew point upper limit of 1.00%, a dust upper limit of 40, and a VOC upper limit of 15. If the dew point of the first region is 0.98, the dust level is 35, and the VOC level is 12, then the first region has no hard limits exceeding the limits. If the dew point of the second region is 1.02, the dust level is 38, and the VOC level is 14, then the second region has a dew point exceeding the limit. If the dew point of the third region is 0.97, the dust level is 42, and the VOC level is 13, then the third region has a dust exceeding the limit. If none of the three items in region D exceed the limit, then region D is normal. The third step is to determine the target environmental parameters exceeding the limit in each sub-region as abnormal states, and then perform a Boolean logic OR operation on all abnormal states.

[0061] For ease of explanation, the dew point exceeding the limit in the second region can be recorded as 1, the dust exceeding the limit in the third region as 1, and the rest as 0. The system performs an OR operation on all abnormal items, and generates a global macroscopic abnormality indication signal as long as at least one item is 1; only when all results are 0 is a global macroscopic normal indication signal generated. Thus, the dashboard only needs to express a macroscopic conclusion, instead of requiring operators to continuously interpret every probability detail. This reconstruction is actually a state degradation expression strategy. When complex trend recognition leads to excessive personnel burden, the system automatically reverts to a more easily understood rule-based expression. Although it reduces the warning lead time, it can significantly reduce the risk of personnel completely ignoring the dashboard due to information overload.

[0062] In abnormal situations, if a key parameter is missing in a certain sub-region, the sub-region can be handled in two ways: First, if other parameters in the same region have already exceeded the limit, it can still be directly judged as abnormal; Second, if there are no hard limits in the same region and the key parameter is missing, the sub-region can be marked as pending review and a maintenance prompt can be attached to the macro signal; If there is no valid data in all sub-regions, the system should not output global normal, but should output status unknown / communication pending inspection signal to prevent data interruption from misleading the field.

[0063] In a specific work shift or the second work shift in the aforementioned injection workshop, the team's signal load has been overloaded and reached the preset overload threshold. At this time, the original probability-based dashboard displays 0.74 in the second area, 0.71 in the third area, and 0.67 in the D area, with a dashboard update frequency of 1 time / min. After the system determines that the fatigue index exceeds the threshold, it no longer displays these probability values, but directly checks whether each area has actually exceeded the safety hard boundary. If only the dew point in the second area reaches 1.02 and the dust in the third area reaches 42, the dashboard will uniformly display that the workshop is in an abnormal state. Please inspect the second and third areas, and at the same time, block the fluctuation values ​​of underlying parameters such as 0.74 and 0.71.

[0064] The purpose of this mechanism is to retain the most critical and actionable security information during high-load phases, thereby enabling the Kanban presentation to adaptively switch from high information density to high comprehensibility.

[0065] In this embodiment, the alarm execution module includes: a first driving unit, used to drive the status indicator board to perform alternating high-frequency flashing display of multiple regions and multiple parameters in response to the target status indication signal being an initial probability warning signal; and a second driving unit, used to drive the status indicator board to perform a uniform constant-on display of the entire region in response to the target status indication signal being a global macro status indication signal, and to shield the display of fluctuation values ​​of the underlying parameters.

[0066] This embodiment provides a dual-mode driving mechanism for the Kanban output layer. Specifically, the aforementioned steps have determined the type of target status indication signal. However, if the display behavior is not synchronously reconstructed, and the signal is changed only in the background while the front-end style remains unchanged, the operator will still be subject to complex flickering interference. Therefore, this embodiment further limits the two types of driving methods for the Kanban.

[0067] Specifically, the first driving unit corresponds to the fine-grained early warning mode. When the target status indication signal comes from the initial probability early warning signal, the signboard uses alternating high-frequency flashing display with multiple areas and parameters. For example, the first area displays dew point with a yellow flashing display, the second area displays VOC with a red flashing display, and the third area displays dust with a yellow flashing display. At the same time, the probability values ​​of 0.63, 0.78, and 0.59 are scrolled on the area borders. The high frequency here does not have to be fixed and can be set within the allowed preset visual recognition frequency range. For example, the risk parameters are refreshed every 2 seconds, and the area list is rotated every 5 seconds to enhance the visibility of early warnings. The second driving unit corresponds to the macro state mode. When the target status indication signal becomes the global macro state indication signal, the signboard switches to a uniform constant-on display across the entire area. The so-called uniform constant-on means that the entire display panel or the entire controlled area uses the same color, the same brightness, and the same stable state. For example, a full-screen yellow constant-on indicates that inspection is required, and a full-screen red constant-on indicates that the hard threshold has been reached.

[0068] At the same time, the display of fluctuation values ​​of underlying parameters is blocked, and dynamic probabilities such as 0.67 and 0.71 are no longer scrolled, nor are individual cells flashed locally. The difference here is not just a different interface display style, but a different driving logic. The former aims to present local trends as early as possible, which is suitable for the stage when the user's attention is still sufficient. The latter emphasizes reducing visual feature conflicts, which is suitable for the stage when the user's response is already close to slow. By separating the driving unit, the controller can directly switch the output strategy according to the target signal without the need to temporarily splice display rules.

[0069] In abnormal situations, if a local display unit on the signboard is damaged, the display priority of its adjacent areas can be increased in fine-grained mode, and a regional code prompt can be added to the border; in macro mode, it can degenerate to only output the central main color block and text prompt, ensuring that the most important information can still be conveyed; if communication delay causes the mode switching command to not be delivered on time, the signboard will cache the most recent valid mode locally, and automatically enter a more conservative unified constant-on mode after the timeout, so as to avoid continuing to maintain complex flickering and aggravating visual load.

[0070] During the morning session in the aforementioned workshop, the work was in its initial stage and the interaction response delay was low. Early fluctuations occurred in the second and third zones. The dashboard displayed the risk of 0.76 dew point in the second zone and 0.68 VOC risk in the third zone in an alternating flashing pattern, facilitating point-to-point troubleshooting. In the late afternoon, the system determined that the fatigue index exceeded the threshold, so the controller issued a switching command to the second drive unit. The entire dashboard turned yellow and remained constantly lit, with only the overall workshop alert displayed in the center. The short text in the second / third zone was reviewed first, and all numerical fluctuations stopped.

[0071] The purpose of this mechanism is to ensure that the adaptive state of the backend is truly reflected in the frontend display behavior, thereby achieving stable switching between the same set of monitoring data and two dimensionality reduction expression methods;

[0072] In this embodiment, the system also includes a disaster blocking module, which is used to: obtain the physical confirmation signal of the disaster in the monitoring area through the external fire linkage interface; determine whether the physical confirmation signal of the disaster contains the preset highest-level disaster indicator; if the physical confirmation signal of the disaster contains the highest-level disaster indicator, bypass the adaptive indication module, generate the highest-level evacuation command, and forcibly trigger the external global sound and light alarm device; if the physical confirmation signal of the disaster does not contain the highest-level disaster indicator, maintain the current operating state of the alarm execution module.

[0073] This embodiment provides a blocking and direct-connection mechanism for extreme disaster scenarios. Specifically, the aforementioned adaptive display logic emphasizes balancing complex noise and personnel response load, but this balance is only applicable to trend warning and general anomaly management. Once physically confirmed high-level disaster signs have appeared on site, continuing to consider fatigue index or display downgrade will delay response. Therefore, this embodiment sets up a disaster blocking module to directly take over the alarm link under the highest level of danger.

[0074] Specifically, the disaster prevention module acquires physical confirmation signals of disasters through an external fire alarm linkage interface. These signals can come from flame detectors, independent smoke detectors, fire alarm control panels, explosion suppression units, or other external safety systems. The system identifies and judges the received signals to confirm whether they contain preset highest-level disaster indicators. For example, any of the following can be defined as the highest-level disaster indicator: open flame confirmation, deflagration confirmation, or mandatory evacuation of the entire area. Once this indicator is detected, the system no longer goes through the adaptive indication module or considers whether the fatigue index exceeds the threshold. Instead, it directly generates the highest-level evacuation command and forcibly triggers external full-area audible and visual alarm devices. This bypass is a change in the execution path at the system process level, switching the usual data acquisition—prediction—fatigue assessment—display degradation—visual board execution path to a fast path of physical confirmation—direct evacuation—mandatory audible and visual alarms.

[0075] As a specific example scenario, suppose the current dashboard is in a constant yellow mode due to the protection strategy, and the external fire alarm control panel suddenly sends a signal with the open flame confirmation flag of the injection area set to 1. At this time, even if the background probability warning is 0.55, the system should immediately upgrade to the highest level alarm across the entire area, activating the red high-frequency flashing warning, alarm output, evacuation arrow linkage, and security access control linkage, instead of first reverting to probability mode or waiting for manual confirmation. If the physical disaster confirmation signal does not include the highest level disaster flag, such as only a single point of smoke awaiting verification or a normal fire alarm interface heartbeat, the current operating state of the alarm execution module should be maintained without interrupting the existing adaptive display process. This can avoid frequent interference from low-level alarms or test signals from the external interface with the normal interactive management logic.

[0076] In an optional anomaly handling mechanism, if the fire alarm linkage interface itself is interrupted, the system should not automatically generate the highest level evacuation command, but should report the interface status anomaly separately to the maintenance terminal; if the combination of exceeding the hard threshold of environmental parameters and interface communication failure occurs at the same time, a secondary enhanced alarm can be triggered, such as requiring the nearest shift leader to verify; if conflicting external signals are received, such as one end reporting open flame confirmation and the other end reporting test mode, the highest level disaster indicator will take priority to ensure that the handling is biased towards safety and conservatism.

[0077] During the aforementioned specific work shift or second work shift in the workshop, the second area has entered a macro-alarm mode due to abnormal humidity and rising VOCs. After a preset time interval, the independent flame detectors and fire control panel near the injection station simultaneously report physical confirmation signals. The system immediately interrupts the current yellow constant-on strategy, directly issues the highest-level evacuation command, forces the plant's audible and visual alarms to start, and switches the evacuation signboard to the safety evacuation instruction interface. The work team can evacuate according to the predetermined route without having to understand the complex meaning of the parameters.

[0078] The purpose of this mechanism is to set an insurmountable safety boundary for the dimension reduction adaptive system, thereby achieving system-level forced switching protection when a real disaster has been physically confirmed.

[0079] In this embodiment, the data acquisition module further includes a data compensation unit, which is used to: monitor the data packet loss rate during the acquisition of multi-source environmental state data; determine the relationship between the data packet loss rate and a preset packet loss threshold and a preset distortion upper limit; if the data packet loss rate is higher than the preset packet loss threshold but not higher than the preset distortion upper limit, then a time-series interpolation algorithm based on linear calculation of adjacent normal data points is used to reconstruct the missing data in the multi-source environmental state data, and the reconstructed data is used as multi-source environmental state data input to the early warning generation module; if the data packet loss rate is higher than the preset distortion upper limit, then forward hold or first-order gradient extrapolation interpolation is used to reconstruct the missing data in the multi-source environmental state data, and the reconstructed data is used as multi-source environmental state data input to the early warning generation module.

[0080] This embodiment provides a data compensation mechanism for communication jitter scenarios. Specifically, the aforementioned scheme assumes that sensor data can arrive relatively continuously. However, in a battery workshop that exceeds the preset spatial dimension, heavy equipment, electromagnetic interference, and wireless link congestion may all lead to data packet loss. If an incomplete sequence is directly sent into the early warning model, it may cause prediction deviations and also cause distortion in the fatigue index calculation. Therefore, this embodiment adds a data compensation unit at the acquisition end.

[0081] Specifically, the system monitors the data packet loss rate of each sensor channel. Assuming a VOC node should theoretically upload 10 sampling points per minute, but actually receives 8, the packet loss rate for that minute is 20%. If the preset packet loss threshold is 15%, the node enters the compensation process. If another dew point node actually receives 9 points, with a packet loss rate of 10%, it continues processing directly based on the original data. For nodes entering the compensation process, the system uses a temporal interpolation algorithm based on linear calculations of adjacent normal data points to reconstruct the missing data.

[0082] For example, the expected data for a certain dust node at consecutive times t1, t2, t3, t4, and t5 are 30, missing, 34, missing, and 38, respectively. Since t2 is between 30 and 34, it can be linearly interpolated to obtain 32; t4 is between 34 and 38, and it can be interpolated to obtain 36. The reconstructed sequence becomes 30, 32, 34, 36, and 38, which is then used as multi-source environmental status data input to the early warning generation module.

[0083] As another specific example, if the dew point sequence is 0.94 at 10:00:01, missing at 10:00:02 and 10:00:03, and 1.00 at 10:00:04, then the two missing points can be reconstructed as 0.96 and 0.98 using equidistant linear interpolation. This maintains the continuity of the time series without artificially introducing abrupt changes. It should be emphasized that the above algorithm for bilateral linear reconstruction based on adjacent normal data points relies on subsequent actual data collection. Therefore, the data compensation unit is equipped with a short-term online delay buffer queue. When a missing data point is detected in real time, the system does not immediately output the gap to the early warning generation module. Instead, it briefly suspends the operation pointer at that moment in the buffer queue, waiting for the nearest normal sampling point, i.e., 10:00:04 mentioned above, to arrive before performing retrospective interpolation to complete the data. Then, the complete time slice data is pushed downstream. Since this buffer time is generally only on the order of seconds, much shorter than the environmental anomaly evolution cycle, it will not disrupt the temporal consistency of the early warning system.

[0084] The reason for using linear interpolation is that in most short-term packet loss scenarios, the actual changes in environmental parameters are usually continuous. Reconstructing by connecting nearby points can obtain approximate values ​​with confidence levels that meet preset requirements. Of course, this method is not intended to permanently replace the actual measurement, but to provide computable input for the prediction model within the communication instability window and prevent the early warning link from being interrupted.

[0085] In abnormal situations, if the waiting timeout buffer queue still does not receive subsequent normal points, or if the packet loss rate is too high and exceeds a higher level of distortion limit, where the higher level of distortion limit refers to the limit of packet loss that causes the confidence of the warning result output by the system based on the interpolated reconstructed data to be lower than the preset safety baseline, for example, reaching more than 50%, then simple bilateral linear interpolation is no longer reliable or cannot be performed. At this time, the system can degenerate into forward hold or first-order gradient extrapolation interpolation using the preceding normal data to ensure that the warning link is not blocked, and label the node as low confidence and downweight the regional results.

[0086] During specific work shifts or the second work shift in the aforementioned workshop, multiple material handling equipment and high-power equipment operate simultaneously, causing short-term congestion in the wireless gateway of the third area. Within 1 minute, two sampling points of VOC data are lost, with a packet loss rate of 20%. The data compensation unit detects that this value is higher than the threshold of 15%, and then suspends the operation for several seconds. After the subsequent points arrive, the two missing values ​​are reconstructed based on the normal points before and after, so that the time series curve of the third area remains continuous. In this way, the early warning model can still correctly identify whether there is a continuous upward trend in the third area, without causing calculation deviations due to the two missing points or being mistaken for an instantaneous recovery to normal.

[0087] The purpose of this mechanism is to maintain the continuity and computability of the monitoring link in unstable industrial communication scenarios, thereby ensuring steady-state input in high-noise environments.

[0088] In this embodiment, the system further includes a threshold update module, which is used to: obtain missed detection feedback records from external operator terminals, and calculate the actual missed detection rate of the status indicator dashboard during the output of the global macro status indicator signal based on the missed detection feedback records; determine the relationship between the actual missed detection rate and the preset missed detection tolerance; if the actual missed detection rate is higher than the preset missed detection tolerance, then lower the preset fatigue threshold; if the actual missed detection rate is not higher than the preset missed detection tolerance, then maintain the preset fatigue threshold.

[0089] This embodiment provides an update mechanism for closed-loop correction of fatigue threshold. Specifically, the aforementioned display degradation can reduce personnel workload, but if degradation is triggered too early or too frequently, it may cause some local anomalies that could have been detected in advance through fine-grained mode to be folded into the macro display, thereby increasing the risk of missed detection. Therefore, this embodiment uses operator terminal feedback to correct fatigue threshold in a closed loop.

[0090] Specifically, the threshold update module obtains missed report feedback records from external operator terminals. These missed report feedbacks can be submitted by team leaders, inspectors, or safety officers on the terminal. The content should include at least the time of occurrence, the corresponding area, the actual anomaly type found on site, and the mode of the dashboard at that time. The system focuses on filtering missed events that occur during the output of global macro status indication signals to assess whether the necessary early warning capabilities have been lost due to excessive dimensionality reduction.

[0091] For ease of explanation, let's assume that the system was in macro mode 40 times in the past week. In 3 of these instances, operators reported that the dashboard only displayed a global alert but did not significantly indicate a continuous increase in VOCs in zone D, and on-site inspections revealed that it was approaching the process limit. The actual false negative rate can be roughly recorded as 3 / 40, or 7.5%. If the preset false negative tolerance is 5%, it means that the system has sacrificed necessary local early warning capabilities due to excessive dimensionality reduction, easily leading to a mode switching frequency higher than the preset standard. In this case, the module executes an action to increase the degradation trigger threshold and lowers the preset fatigue threshold according to preset logic, for example, from 0.65 to 0.60.

[0092] It is important to note that simply lowering the threshold, while logically making it easier to meet the dimensionality reduction condition that the alarm fatigue index is not lower than the fatigue threshold, is not directly related to this. To overcome this inverse mathematical constraint and achieve the practical goal of reducing degradation, the system incorporates a depth index scaling mechanism that is linked to the lowering action while lowering the preset fatigue threshold. That is, when the instruction to lower the threshold is triggered, the system simultaneously applies a depth attenuation mapping to the alarm fatigue index subsequently calculated by the fatigue assessment module, for example, by multiplying the original index by a preset gain adjustment coefficient much smaller than 1, such as 0.5. In this way, although the benchmark threshold for comparison is reduced to 0.60, the alarm fatigue index itself participating in the comparison is significantly compressed.

[0093] For example, the original high-load state of 0.65 is reduced to 0.325 after mapping. This means that the fatigue index after mapping will only exceed the new threshold of 0.60 when the fatigue level calculated in the original calculation reaches an extreme level, i.e., when the original value before mapping exceeds 1.20. Conversely, if the actual false alarm rate is only 2% after a week of statistics, which does not exceed the tolerance, the preset fatigue threshold remains unchanged, and the above-mentioned deep attenuation mapping is not triggered. This preserves the existing human-machine interaction protection capabilities and avoids pushing operators back into complex flashing environments in pursuit of earlier warnings. The reduction of the fatigue threshold here, through the tightly coupled index depth scaling mechanism, actually achieves the technical effect of improving the judgment benchmark for entering the macro mode at the underlying business logic of the system.

[0094] Through this feedback loop, the system of matching thresholds and indices is no longer a fixed constant, but rather undergoes closed-loop dynamic calibration based on on-site missed report feedback. In abnormal situations, if there are too few missed report feedback records in a certain statistical period, such as below the minimum sample size, it is not advisable to update the threshold rashly. Instead, the previous threshold can be used and the observation period extended. If there are obvious conflicts in the feedback records, such as both missed reports and normal prompts with sufficient contradictory evaluations in the same period, the safety administrator can manually review them before proceeding with the calculation. If the operator terminal is offline for a long time, the module will pause automatic updates and only retain logs to prevent threshold deviations caused by incomplete feedback.

[0095] After two weeks of continuous operation in the aforementioned workshop, the system found that the team's aversion to complex flickering had been significantly alleviated. However, night shift inspectors repeatedly reported that in macro mode, VOCs in certain edge areas were rising at a rate lower than the preset gradient without receiving sufficient attention. Although no accidents occurred, this increased the workload of manual investigation. The module's statistics showed a false negative rate of 7.5% during macro mode, exceeding the 5% tolerance. Therefore, an update mechanism was implemented, including lowering the fatigue threshold and initiating deep decay mapping. Subsequently, the system required a higher initial fatigue accumulation to cross the new threshold line after mapping, thus entering macro mode later and retaining more fine-grained warning time.

[0096] The purpose of this mechanism is to enable the alarm strategy to not only adapt to the workload of personnel, but also to perform closed-loop constraints and indirect corrections based on the feedback of missed alarms, thereby achieving a dynamic balance between security and understandability.

[0097] In this embodiment, the monitoring area is the battery production workshop; the multi-source environmental status data includes cross-environmental parameters collected by dew point sensors, dust sensors and volatile gas sensors distributed in the battery production workshop; the historical alarm interaction data comes from the feedback operation records of the battery production workshop operators on the status indicator dashboard.

[0098] This embodiment provides a deployment method for battery production workshops. Specifically, the aforementioned implementation methods can be applied to various industrial environments, but in this embodiment, the monitoring area is clearly limited to the battery production workshop, and a cross-monitoring system is constructed around the three most sensitive parameters in battery manufacturing: dew point, dust, and volatile gases.

[0099] Specifically, dew point sensors can be deployed at the entrance of the drying room, above the liquid injection area, in the material buffer zone, and near the air conditioning supply and return air ducts to capture moisture load changes below a preset value; dust sensors can be deployed in the electrode processing, cutting and stacking transition area, and material flow channel to monitor dust accumulation and the risk of airflow entanglement; volatile gas sensors can be deployed near the liquid injection station, the pre-formation temporary storage area, and the exhaust vent to identify gas changes caused by electrolyte evaporation; the environmental parameters generated by these three types of sensors are not used in isolation, but rather participate in the judgment as cross-environmental parameters.

[0100] For example, when a single dew point rises slightly, the system may not immediately identify it as high risk; however, if VOC increases and dust also shows abnormal fluctuations in the same sub-region, it is more likely to indicate changes in airflow organization, local external infiltration, or enhanced material volatilization. In this case, the prediction model and subsequent dimensionality reduction strategy will respond more quickly.

[0101] Conversely, if only a single dust sensor shows fluctuations while dew point and VOC remain stable, it is more likely due to localized operational disturbances, and the system can avoid over-amplifying them. Regarding historical alarm interaction data, its source is clearly limited to operator feedback records on the status indicator dashboard. This includes confirmation button operations, mute button operations, workstation terminal resets, shift supervisor handheld terminal inspection feedback, missed alarm reports, and manual evaluations after mode switching. By directly using these interactive behaviors as system inputs, alarm fatigue index, threshold updates, and display mode switching are correlated with actual workshop interactions.

[0102] In abnormal situations, such as during partial renovations of the workshop or sensor maintenance, the sensor deployment density in certain areas may temporarily decrease. In this case, the system can operate as existing effective nodes, but a coverage reduction marker will be added to the area-level output to avoid equating low-coverage areas with normal coverage areas. If the communication of a certain type of sensor is temporarily interrupted, such as during VOC system maintenance, the system can still operate based on the remaining dew point and dust data, but the relevant warning confidence level should be automatically reduced and maintenance personnel should be notified.

[0103] In a lithium battery manufacturing plant, the drying room is adjacent to the liquid injection area, and the surrounding area is experiencing continuous humid weather. The dew point sensor near the logistics gate showed a slow overall increase in value, while the VOC sensor near the liquid injection station showed a slight increase. The dust data from the wafer stacking and transfer also fluctuated briefly. The system combines these three types of cross parameters. First, the predictive model generates a probability warning for the second and third areas. Then, combined with the operator's confirmation, silence, and missed report feedback on the dashboard, the system dynamically decides whether to continue maintaining fine-grained display or switch to global macro indication. The entire process revolves around the actual risk mechanism of the battery production workshop.

[0104] The purpose of this mechanism is to embed the aforementioned adaptive alarm and status indication technologies into the real parameter system and operation loop in the battery production scenario, thereby achieving engineering adaptation to the highly sensitive manufacturing environment.

[0105] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An adaptive environmental alarm and status indication system for a battery production workshop, characterized in that, include: The data acquisition module is used to acquire multi-source environmental status data with timestamps and regional location identifiers for the monitored area, as well as historical alarm interaction data; The early warning generation module is used to generate an initial probability early warning signal based on the multi-source environmental state data and through a preset anomaly prediction model. The fatigue assessment module is used to calculate the alarm fatigue index based on the triggering frequency of the initial probability warning signal within a preset time period and the historical alarm interaction data. An adaptive indication module is used to determine the relationship between the alarm fatigue index and a preset fatigue threshold. If the alarm fatigue index is lower than the preset fatigue threshold, the initial probability warning signal is used as the target state indication signal. If the alarm fatigue index is not lower than the preset fatigue threshold, the initial probability warning signal is subjected to dimensionality reduction and reconstruction processing to generate a global macroscopic state indication signal, and the global macroscopic state indication signal is used as the target state indication signal. The alarm execution module is used to send control commands to an external status indicator panel based on the target status indication signal, so as to control the display status of the status indicator panel; The early warning generation module includes: a noise filtering unit, used to filter the multi-source environmental state data based on a dynamic baseline drift algorithm that calculates the mean within a local time window and performs mean removal processing, to generate high signal-to-noise ratio environmental data; and a probability prediction unit, used to input the high signal-to-noise ratio environmental data as a deep reinforcement learning model of the preset anomaly prediction model, and output the initial probability early warning signal, wherein the initial probability early warning signal carries a state fluctuation probability value. The fatigue assessment module is specifically used for: extracting historical alarm confirmation delay time and historical mute operation frequency from the historical alarm interaction data; calculating the signal jump rate of the initial probability warning signal within the preset time period; normalizing the historical alarm confirmation delay time, the historical mute operation frequency, and the signal jump rate; and weighting and summing the normalized historical alarm confirmation delay time, the historical mute operation frequency, and the signal jump rate to calculate the alarm fatigue index. When the adaptive indication module performs dimensionality reduction and reconstruction processing on the initial probability warning signal, it specifically performs the following: masking the state fluctuation probability value in the initial probability warning signal; dividing the monitoring area into a preset number of sub-regions and extracting target environmental parameters that exceed a preset safety hard threshold from the multi-source environmental state data in the sub-regions; determining the target environmental parameters as abnormal states and performing Boolean logic OR operations on all target environmental parameters in abnormal states to generate the global macroscopic state indication signal.

2. The adaptive environmental alarm and status indication system for a battery production workshop according to claim 1, characterized in that, The alarm execution module includes: a first driving unit, used to drive the status indicator board to perform alternating high-frequency flashing display of multiple regions and multiple parameters in response to the target status indication signal being the initial probability warning signal; and a second driving unit, used to drive the status indicator board to perform uniform constant-on display of the entire region in response to the target status indication signal being the global macro status indication signal, and to shield the display of fluctuation values ​​of underlying parameters.

3. The adaptive environmental alarm and status indication system for a battery production workshop according to claim 1, characterized in that, The system also includes a disaster blocking module, which is used to: acquire a physical disaster confirmation signal of the monitoring area through an external fire alarm linkage interface; determine whether the physical disaster confirmation signal contains a preset highest-level disaster identifier; if the physical disaster confirmation signal contains the highest-level disaster identifier, bypass the adaptive indication module, generate a highest-level evacuation command, and forcibly trigger the external global audible and visual alarm device; if the physical disaster confirmation signal does not contain the highest-level disaster identifier, maintain the current operating state of the alarm execution module.

4. The adaptive environmental alarm and status indication system for a battery production workshop according to claim 1, characterized in that, The data acquisition module further includes a data compensation unit, which is used to: monitor the data packet loss rate during the acquisition of the multi-source environmental status data; and determine the relationship between the data packet loss rate and a preset packet loss threshold and a preset distortion upper limit. If the data packet loss rate is higher than the preset packet loss threshold but not higher than the preset distortion upper limit, then the missing data in the multi-source environmental state data is reconstructed using a time-series interpolation algorithm based on linear calculation of adjacent normal data points, and the reconstructed data is input into the early warning generation module as the multi-source environmental state data. If the data packet loss rate is higher than the preset distortion limit, the missing data in the multi-source environmental state data is reconstructed by using the preceding normal data for forward hold or first-order gradient extrapolation interpolation, and the reconstructed data is input into the early warning generation module as the multi-source environmental state data.

5. The adaptive environmental alarm and status indication system for a battery production workshop according to claim 1, characterized in that, The system also includes a threshold update module, which is used to: obtain missed detection feedback records from external operator terminals, and calculate the actual missed detection rate of the status indicator dashboard during the output of the global macro status indicator signal based on the missed detection feedback records; determine the relationship between the actual missed detection rate and a preset missed detection tolerance; if the actual missed detection rate is higher than the preset missed detection tolerance, then lower the preset fatigue threshold. If the actual false negative rate is not higher than the preset false negative tolerance, then the preset fatigue threshold is maintained.

6. An adaptive environmental alarm and status indication system for a battery production workshop according to any one of claims 1 to 5, characterized in that: The monitoring area is the battery production workshop; the multi-source environmental status data includes cross-environmental parameters collected by dew point sensors, dust sensors and volatile gas sensors distributed in the battery production workshop; the historical alarm interaction data comes from the feedback operation records of the battery production workshop operators on the status indicator dashboard.

Images