Digital Input Edge Detection Using Intelligent Filtering Algorithm
Through the digital input edge detection method with intelligent filtering function, the root causes of systemic failures and accidents are identified and analyzed, and the difficulties of identifying multiple accident association analysis in the existing technology are solved, achieving more accurate and efficient incident reporting and fault analysis , enhances the understanding of systemic failures.
Patent Information
- Application Number
- CN202010520708.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2020-05-31
- Filing Date
- 2020-06-09
- Publication Date
- 2024-12-31
- Estimated Expiration
- 2040-06-09
AI Technical Summary
It is difficult for the prior art to effectively identify and analyze the root causes of systemic failures and accidents, especially in the analysis of correlations between multiple accidents, resulting in poor results in long-term corrective measures.
Using digital input edge detection methods and systems with intelligent filtering function, through digital filters and state edge detectors, it ignores signal changes caused by self-testing, contact jumps and external interference, and provides a more accurate sequence of event analysis to determine faults. The main reason. The configurable time constant of the digital filter and the integrated de-hop counter improve analysis flexibility and accuracy.
Effectively identify and record event sequences, reduce the overhead of CPU resources, improve the accuracy and response speed of event reports, and accurately identify singular events when the actual state changes, reducing false alarms caused by self-testing and external influences. , enhances the understanding of systemic failures and causes of accidents.
Smart Images
Figure CN112131030B_ABST
Abstract
Description
Technical Field
[0001] Embodiments of the present invention generally relate to data processing and root cause analysis. Embodiments also relate to identifying root causes of systemic failures and incidents in a work environment where digital filters and edge detectors are used to specify events associated with incidents occurring during operation. Background Art
[0002] Work environments (such as process control, nuclear power, health care, military and manufacturing) often lack a common understanding of the systematic failures and causes associated with accidents. An "accident" in these examples can be an "actual loss" event that can be investigated so that effective actions can be taken to prevent the same or similar losses from happening again. A "near error" event or abnormal condition can also be considered an "accident". In this case, actual losses may not have occurred, but potential losses may have existed. A "near error" can be considered as a situation where, if the situation is slightly different, "actual losses" (e.g., property damage, environmental impact, or human loss) or operational interruption may have occurred. Therefore, such events can be considered as "accidents", "process upsets" or "system upsets", where such "systems" include artificial components and machine parts of the system in the work environment.
[0003] Accidents can be caused by failures of field equipment (such as instruments, control valves, and pumps) or by some form of process disturbance that causes the operation to deviate from the normal operating state. Accidents can also occur due to human factors and related failures, such as human errors or problems with the current management system (e.g., training, work processes, procedures, communications, supervision, etc.). Accidents can also be caused by unexpected external causes. Early detection of such failures, whether they are operating practice failures, equipment failures, or management system problems, enables the operating team to improve current operations to prevent future accidents or support with more effective responses when problems occur. Abnormal conditions can evolve from operational upsets, which can become catastrophic events involving serious damage and harm to the plant and the surrounding community. Continuously learning from accidents and understanding systemic failures are the key to forming effective corrective measures with long-term beneficial effects. Safety management systems can detect and record events that indicate deviations from normal process operations. Some examples of such events include changes in the state of valve limit switches, steam pressure exceeding limits, maintenance overrides or forced points, field failures, or failures in safety manager input / output interfaces. The scan time of the safety management system for the event sequence will typically be equal to the application cycle time, and for low latency sequences of event management, the scan time can be set to 1 millisecond (ms). Each process quantity that changes is an event recorded in an event buffer that resides within the system. The event sequence function records the event if an event occurs for a point that applies to both a process point and a system point. Or a force is applied or removed at a certain point. The event record contains some identification of the event sequence, such as a description of the event as described in the associated event description, as identified by a specific time point or other unique designation. All recorded events are sent to a management system that collects and manages such information.
[0004] Based on the foregoing, it is believed that there exists a need for improved methods and systems for effectively identifying the root causes of systemic failures and incidents. Specifically, failures and root causes need to be understood at two fundamental levels: (1) individual incidents; and (2) more importantly, incident to incident. It is believed that analyzing incident to incident is a key factor in identifying systemic failures and root causes. Otherwise, one will continually solve problems independent of individual incidents, and the likelihood of successful corrective actions having the desired long-term and far-reaching effects will be low.
[0005] Digital filtering of inputs and change of state edge detection to report event sequences when the monitored voltage or current may cause interference from self-test, contact bounce, and other external influences. This also requires hysteresis performance to ignore samples taken when the input signal is traveling between states. The goal is to apply a timestamp where the sample transition actually occurred, while preventing false events from being reported when the sample data is affected by internal or external factors, and avoiding premature reporting of transitions when the input transitions are slow. Summary of the invention
[0006] The present invention provides a method and system for digital input edge detection with intelligent filtering. More specifically, digital filtering of inputs for event reporting sequences and variations in state edge detection are provided that ignore variations in received signals due to interference from self-test, contact bounce, and other external influences. An improved event analysis sequence is provided that is able to better determine the primary cause of a fault sequence. Digital filters are more flexible than hardware filters because the filter time constants can be configurable. BRIEF DESCRIPTION OF THE DRAWINGS
[0007] Figure 1 An I / O board and a processing board are shown using digital filters to detect events.
[0008] Figure 2 The digital input is shown over time to illustrate the effect of implementing a digital filter. DETAILED DESCRIPTION
[0009] Event sequences are used in many industries to identify fault sequences. In the case of a shutdown or failure, a potentially large number of data points are collected. These data points include input state changes with high-resolution timestamps when state changes occur. Engineers later review the logs to determine the main causes of the failure (such as bearing shaking, incoming power abnormalities, loss of A / C synchronization with the rest of the network, loss of steam pressure, or steam temperature or other temperature changes). These main causes of shutdown or failure can vary significantly depending on the specific industry. Some of these reasons require equipment maintenance, and some external reasons may even involve monetary losses to suppliers. The sequence of event detection and reporting is usually completed by sampling and analyzing the input data for "state changes" at a high rate, and when the sequence of event detection and reporting is detected, a high-resolution timestamp is applied using a real-time clock source synchronized between all devices.
[0010] The present disclosure relates specifically to low-latency sequencing of events. In a system, there may be hundreds or thousands of channels transmitting information. In some systems (such as the exemplary system), there are 32 channels that are continuously monitored, although in reality monitoring occurs frequently (e.g., every millisecond). Most importantly, the events are recorded in the order in which they occurred. This can generate a large amount of data. When there is an incident, the sequence of events can be returned and analyzed and the exact time can be determined even down to the millisecond to determine when the event occurred. This can allow the operator to subsequently determine the root cause of the abnormality or simply determine that the operation is normal.
[0011] In a typical system, there may be limitations on the system's ability to process data (such as limited speed on a processor or limited memory), which makes it important to develop an effective method for analyzing data. 32-bit input / output modules are utilized in the disclosure described herein. About 40 to 64 of these modules are connected to a controller. The module performs all sampling on inputs and edge detection. The sampled information is then analyzed to determine the sequence of events.
[0012] When analyzing event information, it is important to monitor when an edge or actual change occurs. For example, a particular reading is constantly "on" and then switches to "off". It is necessary to determine whether this is an actual change or a one-time occurrence, in which a particular reading switches to "off" but then immediately returns to its previous state. In addition, continuous tests also occur, which need to be distinguished from the real situation that needs to be recorded. Therefore, it is necessary to record real events and remove self-tests. In addition, there are situations that can contain "jumps" in the signal. In some cases, there is no obvious moment of "on" in time. Instead, there are short-term fluctuations before steady-state operation. When monitoring these situations where there are jumps or fluctuations, it is necessary to take this into account and be able to ignore or filter out insignificant events. There is a need to be able to ignore the way that jumps occur within a limited time period. In addition, there are usually self-tests that occur that should not be included in the event sequence analysis. Then, the goal is to be able to identify singular events when a state change actually occurs. The system that monitors each of the channels will read whether the signal is "on" or "off", or "1" or "0". This type of monitoring requires a minimum amount of processing capacity.
[0013] In the present disclosure, one way to monitor a signal is to read a voltage which may be a gradual or instantaneous change. A given voltage, for example, exceeding 5 volts may be read as "on" or when it is below 1 volt as "off." An "edge" is defined as a point in time when a voltage is above a specified voltage, such as when the voltage exceeds a 5 volt threshold.
[0014] In safety systems, devices that perform safety functions must maintain their safety integrity level, which includes continuous self-testing and diagnostics of their input circuits. These self-tests can temporarily affect the results of input sensing and affect the sample data used for event reporting sequences. In order to provide accurate sequencing of events while performing self-tests to maintain safety integrity, reliable methods are needed to detect state changes while filtering out invalid sample data, including self-test data.
[0015] An action that can be taken in developing accurate sequencing of events is to perform filtering to prevent external influences of discrete components. However, this can result in an unchanging time constant, providing a slower response time under all conditions, and is susceptible to component failures, tolerances, or drifts. The event data collection sequence can be stopped when performing self-tests. This will mean that auxiliary methods are still needed to manage detected but unplanned factors that affect the input, such as the loss of redundant equipment. In addition to or instead of using a counting filter, the method can be integrated with a de-bounce counter. When two comparator readings indicate that the input has changed to the opposite state, the de-bounce counter will start the counter. At any time when the value crosses the transition of the two comparator thresholds, the counter will reset. When the counter reaches its target value, the transition will be reported. In some cases, this can provide a more accurate timestamp of the leading edge of the transition. In other cases, inputs with a small amount of filtering and significant contact bounce may take longer to resolve and are less accurate because the counter remains reset.
[0016] Now we have developed Figure 1 A better approach is shown in which there is a digital filter (100) and edge detector (101) enhanced by integrating a method for excluding known invalid samples and a method for excluding samples taken when the input is in a travel range. Technical benefits of the present invention include greater flexibility of the digital filter (100) compared to hardware filters, which can be implemented in, for example, software or programmable logic because the filter time constant can be configurable. By using the exclusion mask as an input to the filter, the engine (102) that collects input samples continues to execute while the device performs a self-test and a continuous stream of sample data can be fed to the filter. Compared to the prior art, the method better approximates the start of a transition by maintaining a history of samples taken before the exclusion range. The applicant of the present invention is responsible for marking sample ranges during the execution of the self-test and when the input value is known to be invalid.
[0017] The present disclosure adds a low latency sequence of event reporting to the safety I / O module. The module periodically runs diagnostics that briefly affect the input value. The module supports up to 32 channels of digital input with low latency SOE, which introduces CPU performance risks. One advantage of this filtering method is that it reduces the overhead of managing the SOE state machine on the CPU and allows it to focus on performing safety functions.
[0018] The process is characterized by its reliance on periodic transitions in voltage at the input point. These samples (103) must be taken at a fixed frequency to provide known time differences between data points. The sampled values (103) are compared (104) to two sets of configurable thresholds (105) to provide a set of high and low comparison results. For each sample, a valid bit is maintained to indicate whether the sample has been subject to the self-test and should be rejected when fed into the filter (100).
[0019] like Figure 1 The filter algorithm (100) shown has a programmable filter time value to allow flexibility based on the application. The filter also uses an initial state and filter count to allow detection to start in a high state or a low state. If the sampling and filtering are implemented on different devices connected by a communication link (107), the filter algorithm (100) takes a 16-bit word for each of the high and low comparison values (106a) (or 106b). The communication link can be any communication link (in Figure 1 SPI link is depicted in ) and a 16-bit exclusion mask. The filtering algorithm (100) iterates through each bit and performs the following operations: If the sample is valid and the input is detected as high, the filter count (108) is increased. If the sample is valid and the input is detected as low, the filter count (108) is decreased. If the sample is marked as invalid or the state is neither high nor low, the filter count (108) is not changed. If the filter count (108) reaches the highest count and the previous state is low, or if the filter count (108) reaches the lowest count and the previous state is high, the filter (100) outputs an event result to indicate the new state and the number of bits of the sample when the transition is completed (109). In order to allow the program to pause filtering and continue mission critical functions, the filter always presents the current sample counter and the last known state (109) after processing each group of words. When time permits and more physical sample data has been collected, the filter resumes by accepting this count and state as input.
[0020] The main difference between the present disclosure and the prior art is the ability to exclude samples based on a string of valid / invalid samples that have a 1 to 1 correlation with an input sample. This may include an exclusion mask that filters out samples such as those recorded when there is a "jump" condition where the sample exhibits a uniform fluctuation into steady state over a short period of time.
[0021] Figure 1 A system of I / O boards and processing boards that generate and sample signals is shown. The upper left (103) shows samples with channel 1, channel 2, and then the last of the 32 channels, channel 32. In this example, there are 255 samples per channel, with each sample taken 0.256ms apart. After the recording of a bit is at a high level for a given number of bits, for example, after the 20th sample, it will be considered an event to be recorded. You continue to process through the bit array until there is a transition where the state has changed. The analysis continues on the I / O board until a transition is reached. In the lower left box, there is a digital filtering box (100) and a determination of the valid sample array (102).
[0022] Figure 2 A 4 ms filter digital input delay is shown. Specifically, it shows time intervals that would be disregarded as showing random events (200) or other data that does not indicate an event that needs to be included in the sequence of events being recorded. Starting from the top of the figure, a time series of (201) 0.256 milliseconds of each sampling time is shown. Figure 2 The next segment shows the combined input over time (202), with an extended plateau shown on the left portion and multiple irregular fluctuations of the input signal including a test pulse on the far right side shown on the right. The next two segments show the accumulated time (203) when the input signal is at or above the high compare state or the low compare state (204). Then, there is a segment (205) showing the portion of the timeline that is ignored due to the test pulse.
[0023] Specific implementation plan
[0024] While the following is described in conjunction with specific embodiments, it should be understood that this description is intended to illustrate and not to limit the scope of the foregoing description and the appended claims.
[0025] A first embodiment of the present invention is a method for detecting and subsequently identifying abnormal events in a process, the method comprising the steps of obtaining a plurality of samples (103) at a fixed sampling frequency to provide known time differences between process data points; sending the plurality of samples (103) to a digital filter (100); using the digital filter (100) to compare each of the samples with two sets of configurable thresholds (105), wherein the two sets of configurable thresholds are different to provide a set of high and low comparison results; determining the relationship of each of the samples to the configurable thresholds (104); determining whether each of the plurality of samples is repeated; counting the number of valid samples to be counted as a filter count (108); and recording an event (109) when the filter count (108) reaches a pre-specified level. An embodiment of the present invention is one, any one, or all of the previous embodiments in this paragraph to the first embodiment in this paragraph, further comprising determining whether each sample is valid or invalid. An embodiment of the present invention is one, any one, or all of the previous embodiments in this paragraph to the first embodiment in this paragraph, wherein the invalid sample is discarded after determining that the invalid sample is invalid. An embodiment of the present invention is one, any or all of the previous embodiments in this paragraph to the first embodiment in this paragraph, wherein the sample is determined to be invalid based on observing a fluctuating sample value followed by a stable sample value. An embodiment of the present invention is one, any or all of the previous embodiments in this paragraph to the first embodiment in this paragraph, wherein the sample is determined to be invalid based on a sample string having a 1 to 1 correlation with an input sample. An embodiment of the present invention is one, any or all of the previous embodiments in this paragraph to the first embodiment in this paragraph, wherein the filter count (108) includes a predetermined value defined as constituting an event to be recorded. An embodiment of the present invention is one, any or all of the previous embodiments in this paragraph to the first embodiment in this paragraph, wherein the filter count (108) decreases when the process data point is within two sets of configurable thresholds. An embodiment of the present invention is one, any or all of the previous embodiments in this paragraph to the first embodiment in this paragraph, wherein the digital filter (100) adopts an initial state and a filter count (108). An embodiment of the present invention is one, any or all of the previous embodiments in this paragraph to the first embodiment in this paragraph, wherein the event is recorded in an event collector stored in a memory on the processor (109). An embodiment of the present invention is one, any or all of the previous embodiments in this paragraph to the first embodiment in this paragraph, wherein the determination of the relationship is that a plurality of process data points are above a configurable upper threshold value.An embodiment of the invention is one, any or all of the preceding embodiments in this paragraph through the first embodiment in this paragraph, wherein the determination of a relationship is that a plurality of process data points are below a configurable lower threshold.
[0026] Although there is no further detailed description, it is believed that those skilled in the art can utilize the present invention to the greatest extent by using the foregoing description and can easily determine the essential characteristics of the present invention without departing from the spirit and scope of the present invention to make various changes and modifications of the present invention and adapt it to various usages and conditions. Therefore, the aforementioned preferred specific embodiments should be understood as merely illustrative and not to limit the rest of the present disclosure in any way, and are intended to cover various modifications and equivalent arrangements included within the scope of the appended claims.
[0027] In the foregoing, all temperatures are set forth in degrees Celsius and all parts and percentages are by weight unless otherwise indicated.
Claims
1. A method for detecting and subsequently identifying abnormal events in a process, the method comprising the following steps: a. obtaining multiple samples (103) at a fixed sampling frequency to provide known time differences between process data points; b. sending the plurality of samples (103) to a digital filter (100); c. using the digital filter (100) to compare each of the samples to two sets of configurable thresholds (105), wherein the two sets of configurable thresholds are different to provide a set of high and low comparison results; d. determining the relationship of each of the samples compared to the configurable threshold (104); e. determining whether each of the plurality of samples is repeated; f. Counting the number of valid samples to be counted as the filter count (108); g. If the sample is valid and detected as low, the filter count (108) is decreased; if the sample is marked valid and detected as high, the filter count (108) is increased; and h. Record an event when the filter count (108) reaches a pre-specified level.
2. The method of claim 1 further comprising determining whether each sample is valid or invalid. The method of claim 2 , wherein the sample is discarded after determining that the sample is invalid. The method of claim 3 , wherein determining that the sample is invalid is based on observing a fluctuating sample value followed by a stable sample value. The method of claim 3 , wherein determining that the sample is invalid is based on a string of samples having a 1 to 1 correlation with an input sample.
6. The method of claim 1, wherein the filter count (108) comprises predetermined values defined as constituting events to be recorded.
7. The method of claim 1, wherein the filter count (108) is decreased when the process data point is below a lower threshold.
8. The method of claim 1, wherein the digital filter (100) employs an initial state and a filter count (108).
9. The method of claim 1, wherein the events are recorded in an event collector stored in a memory on a processor.
10. The method of claim 1, wherein the determination of the relationship is the plurality of process data points being above a configurable upper threshold or below a configurable lower threshold.