WO2022221944A1

WO2022221944A1 - System and method for cathodic protection monitoring

Info

Publication number: WO2022221944A1
Application number: PCT/CA2022/050598
Authority: WO
Inventors: Antonio Da Costa; Matthew Barrett; William MAIZE
Original assignee: Mobiltex Data Ltd
Priority date: 2021-04-22
Filing date: 2022-04-20
Publication date: 2022-10-27
Also published as: GB2620862A; GB202316052D0; EP4327111A1; CA3217048A1; AU2022259875A1

Abstract

The system and method of the present application improve pipeline integrity and may optimize operational efficiencies. Using the data that is generated by cathodic protection and pipeline integrity monitoring devices (e.g. rectifier monitoring), the present application utilizes data analytics techniques, such as artificial intelligence and machine learning algorithms, to improve pipeline integrity operations.

Description

SYSTEM AND METHOD FOR CATHODIC PROTECTION MONITORING

FIELD

[001] The present application relates to cathodic protection systems and more specifically to improved monitoring of cathodic protection systems using environmental data and specific alarm thresholds.

BACKGROUND

[002] The cathodic protection industry collects, stores and analyzes large amounts of data. A myriad of data points are collected and stored each year for various purposes, for example regulatory compliance purposes, and for tracking operating performance. While the quality and quantity of data has improved over the years, influenced by increasingly automated data collection using remote monitoring devices, most data collected remains unutilized or underutilized (e.g. stored away in a unrelatable database for reference only if a specific need arises).

[003] There is a need to better assess and plan maintenance of cathodic protection systems and pipelines, and also to optimize performance of cathodic systems and pipelines.

[004] Since the introduction of remote monitoring technologies onto cathodic protection assets almost 20-years ago, operators have worked with technology vendors to develop alarms, often simple notifications via email or SMS that are triggered if a threshold has been breached on a pre- configured limit. Alarms ensure that the right people are notified as soon as data, indicative of a potential problem, exists. Alarms help pipeline operators monitor their cathodic protection systems and have saved time and resources in the process.

[005] There are various potential causes of an alarm. Changes in readings measurements can be indicative of both the state of the overall cathodic protection system (e.g. rectifier is supplying an adequate supply of voltage to the pipe), the performance of the cathodic protection system assets (e.g. the rectifier stops working), and the performance of the remote monitoring unit (RMU) in itself (e.g. the RMU fails to communicate). [006] If an alarm limit is poorly configured, it can become the root cause of a very challenging, and sometimes dangerous condition - alarm fatigue. An operator or technician develops alarm fatigue when the volume of alarms generated by the remote monitoring system becomes overwhelming, ceases to compute with known operating conditions, or is based on too many false positives.

[007] When an operator or a technician develops alarm fatigue, the first thing that they may do is to adjust the limits of the device, essentially the threshold value at which a measured reading is higher, or lower (depending if Upper or Lower limit) must surpass to necessitate that an alarm condition is classified, and an alarm notification is circulated. However, this widening of limits, driven only by the desire to reduce the amount of alarms generated, is a dangerous precedent. Limits that are established for the purpose of avoiding an alarm defeats the purpose of remote monitoring systems. For a manufacturer of remote monitoring devices, alarm fatigue contributes to unreliable values that the remote monitoring program has collected. Homogeneity of alarms with respect to the root causes is also a problem.

SUMMARY [008] The system and method of the present application improve pipeline integrity and may optimize operational efficiencies. Specifically focusing on the data that is generated by cathodic protection and pipeline integrity monitoring devices (e.g. rectifier monitoring), the present application utilizes data analytics techniques, such as artificial intelligence and machine learning algorithms, to improve pipeline integrity operations and the safety of workers and the broader public. Examples of artificial intelligence and machine learning work, such as applying intelligent algorithms to data analysis streams, are provided as a means of reducing data overload while providing automated predictive failure analysis and optimization of cathodic protection systems.

[009] Data analytics and machine learning may be used to automate the setting of remote limits and alarm thresholds and to provide flexibility for systemic seasonal variations that may occur throughout the year. Machine learning algorithms may be developed to recognize patterns in historical rectifier readings (e.g. rectifier resistance V/I) in relation to other operation or relatable data - such as soil type, seasonality and precipitation - ultimately providing prescriptive insights into what future thresholds at which alarm limits (e.g. upper and lower alarm limits) should be set to provide an optimal feedback and security layer.

[0010] Alarm optimization may play a vital role in the long-term success of cathodic protection engineers and pipeline operators in keeping their critical assets, and the safety of the public, well protected. Use of data generated by cathodic protection systems may play a useful role in monitoring these systems and pipelines in a more efficient manner. Alarms regarding the state or level of protection that the CP system is achieving are also subject to seasonal and meteorological variations. This diversity enables alarms to be classified into a variety of buckets, each with its own rank and signature in terms of priority and need for remedial action. A factor that greatly contributes to alarm fatigue is that in current user interfaces, cathodic system software management systems and operating systems, alarms are not often adequately classified and sorted, and thus appear as a foreboding wall of homogeneous problems.

[0011] Rectifier monitors, coupon-based monitors, test point monitors, bond monitors and electrical resistance or ultrasonic probes and monitors are programmed with low and high thresholds for each of the measurement channels (e.g. voltage, current, etc.). For example, if the rectifier resistance (e.g. V/I) or any monitored characteristic falls above or below the thresholds, an alarm may be issued.

[0012] A method for managing a cathodic protection system is provided, the method comprising obtaining data from one or more rectifier monitors, obtaining data from one or more test station monitors, transmitting the rectifier data and the test station data to a monitoring system via a communications network, analyzing the rectifier data and the test station data using machine learning algorithms to manage the performance of cathodic protection systems, and transmitting control signals to the one or more rectifiers based on the data analysis and algorithm results.

[0013] A method for managing a cathodic protection system is provided, the method comprising obtaining data from one or more monitoring units; transmitting the data to a monitoring system via a communications network; analyzing the data using machine learning algorithms to manage one or more first system components in the cathodic protection system; and transmitting control signals to one or more second system components based on the data analysis and algorithm results.

[0014] A method for managing a cathodic protection system is provided, the method comprising obtaining data from one or more rectifiers, from one or more coupon test stations and from one or more pipe-to-soil test stations or other remote monitoring and data collection units (e.g. RMUs), transmitting the data to a monitoring system, determining, from the data, limit set points for the one or more rectifiers and updating the set points for the one or more rectifiers. The data may include time-series voltage data and time-series current data. The limit set points may include voltage, current and resistance and derivatives, integrals and other mathematical measures, functions or statistics derived from these. Data analytics and machine learning may be used to determine set points, for example with machine learning algorithms for recognizing patterns in historical rectifier readings. The rectifier set points may be updated and adjusted manually or by signaling the rectifier with over-the-air control (e.g. using a controller device such as RMU3 + control block or control signals from monitoring platform).

[0015] A method for determining the optimal DC output of a rectifier is provided, the method comprising obtaining data from one or more rectifiers, from one or more coupon test stations, and from one or more pipe-to-soil test stations, transmitting the data to a monitoring system, determining from the data an optimal level of one or more operational parameters (e.g. voltage output, current output, resistivity or connectivity of system components) based on the data, and adjusting the rectifier current to achieve the optimal level of the one or more operational parameters. For example, potential is observed at the test stations (e.g. one or more coupon test stations and the one or more pipe-to-soil test stations) and adjusting the rectifier current changes that observed potential. The data may include time-series voltage data and time series current data. The optimal level of DC current is based on regulatory compliance criteria (e.g. 850m V).

[0016] A method for determining optimal DC output of one or more rectifiers is provided, the method comprising obtaining data from the one or more rectifiers, obtaining AC interference data, transmitting the data and AC interference data to a monitoring system, determining from the rectifier data and the AC interference data an optimal level of DC current at the one or more rectifiers, adjusting the DC output of the rectifier based on the determined optimal DC current level. The AC interference data may be obtained from one or more electromagnetic field (EMF) sensors or from AC load data from a power utility. The optimal DC current level may be based on established requirements (e.g. National Association of Corrosion Engineers (NACE)) such as guidelines for balancing DC current density and AC current density on the pipeline. From the one or more EMF sensors on local HVAC lines, or AC load data from a power utility it may be determined the influence of AC interference on the pipeline and coupons.

[0017] A method for generating a classification dataset is provided, the method comprising: obtaining data for one or more rectifiers; fitting the data using a fitting procedure to obtain fit parameters; determining a location for the one or more rectifiers; obtaining environmental data for the one or more rectifier location; classifying the one or more rectifiers based on the environmental data; and generating the dataset comprising data for each of the one or more rectifiers identifying classification and corresponding fit parameters.

[0018] A method for predicting alarm thresholds is provided, the method comprising: determining a location of one or more rectifiers; obtaining environmental data for the location of the one or more rectifiers; classifying the one or more rectifiers based on the environmental data; and generating a predicted alarm threshold based on data statistics that correspond to the classification.

[0019] A method for determining alarm thresholds, the method comprising: obtaining data at one or more rectifiers over a period of time; fitting the data with a fitting procedure to obtain fit parameters; classifying the one or more rectifiers based on the fit parameters; and generating alarm thresholds based on the determined classification.

[0020] A cathodic protection monitoring system, the system comprising one or more monitoring units for collecting data, a monitoring subsystem, for receiving the data over a communications network, the monitoring subsystem comprising a processor configured to: determine, from the data, limit set points for one or more first system components and update, over the communications network, the limit set points for one or more second system components. The monitoring units may include for example remote monitoring units on rectifiers and test stations.

[0021] A cathodic protection monitoring system, the system comprising one or more monitoring units for collecting data, a monitoring subsystem, for receiving the data over a communications network, the monitoring subsystem comprising a processor configured to analyze the data using machine learning algorithms to manage one or more first system components in a cathodic protection system; and transmit, over the communications network, control signals to one or more second system components based on the data analysis and algorithm results. The system may further comprise one or more test station monitors for collecting data, wherein the data is transmitted to the monitoring subsystem.

[0022] To better protect pipeline assets from corrosion, and to ensure that alerts do not become overwhelming, seasonally adjusted alarm limits are determined and provided. In the example method of the present application, this involves two phases. A first phase of predicted thresholds and a second phase of rectifier specific thresholds.

[0023] For predicted thresholds, upon installation of a new remote monitoring unit (RMU) a machine learning algorithm suggests an alert threshold. Set points can also be added to devices so that when a monitor detects movement of a measured characteristics above or below a set point, the operation of the device is correspondingly changed. Set points and thresholds may be statically set, e.g. when a measured value is exceeded an alarm is triggered or device operation is changed, and may also be dynamically set, for example, when the measured value of a characteristics changes by more than 50% an alarm is triggered or device operation is set. The value of 50% is exemplary only of how dynamic thresholds and set points may work and other values may be used. The suggested range may be based on features specific to specific locations along a pipeline such as soil type, weather, type of pipeline, pipe depth and pipeline coating. For example, the suggested range may be based on publicly available soil and temperature classifications specific to the rectifier’s location, as well as features describing the pipeline and cathodic protection system. [0024] Rectifier specific thresholds may be implemented in the weeks and months after an RMU is activated (e.g. after the collection of sufficient readings). This allows rectifier resistance readings to be captured and a location specific dataset is constructed. With the data-based thresholds, the alert threshold becomes more focused and specific to the rectifier being monitored (e.g. specific to the environmental factors at the rectifier’s specific location).

[0025] There is also provided a computer program product comprising a non-transitory computer readable medium having instructions stored thereon, which when executed by a processor, the processor performs any one of the methods described in the present application.

BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 illustrates a corrosion protection system according to an example embodiment of the present application;

[0027] Figures 2A and 2B illustrate a system for cathodic protection monitoring according to an example embodiment of the present application;

[0028] Figure 3 A illustrates a graph showing a weather seasonal variation in DC current of the cathodic protection system according to an example embodiment of the present application;

[0029] Figure 3B illustrates a cosine function fit graph of the seasonal variation of the DC system resistance (voltage/current) according to an example embodiment of the present application;

[0030] Figure 4 illustrates a histogram of phase offset fit parameters according to an example embodiment of the present application; [0031] Figure 5 illustrates rectifier system resistance fit parameters (amplitude and offset) for various temperature classifications according to an example embodiment of the present application;

[0032] Figure 6 illustrates rectifier system resistance fit parameters (amplitude and offset) for various moisture classifications according to an example embodiment of the present application;

[0033] Figure 7 illustrates a decision tree diagram for soil temperature regime classification according to an example embodiment of the present application;

[0034] Figure 8 illustrates a graph showing seasonally adjusted limits define a range of acceptable V/I readings according to an example embodiment of the present application;

[0035] Figure 9 illustrates a SSURGO (Soil SURvey GeOgraphic) map;

[0036] Figure 10 illustrates an example of the various Map Units dividing a mixed use industrial/residential location;

[0037] Figure 11 illustrates a cosine fit to rectifier resistance readings over time according to an example embodiment of the present application;

[0038] Figure 12 illustrates a process for predicting alarm thresholds for a rectifier according to an example embodiment of the present application; and

[0039] Figure 13 illustrates a process for determining location specific alarm thresholds for a rectifier according to an example embodiment of the present application. DETAILED DESCRIPTION

[0040] The present invention relates to cathodic protection monitoring systems and more specifically to utilizing a combination of historic data, environmental features and customer provided values to determine alarm thresholds, limits and operational set points for monitored cathodic protection systems and their interoperable (connected) components, and consequent usage for managing cathodic protection systems and their components, for optimizing the operation of such systems and components, and for planning maintenance, replacement, technician scheduling (e.g. for onsite visits), workflow and deployment and oversight. Limit set points may include an indication that a system component is disconnected or inoperable.

[0041] Figure 1 illustrates an exemplary corrosion protection system 100 according to an example embodiment of the present application. As shown in Figure 1, a pipeline 14 is buried in soil 12. The buried pipeline 14 is connected by a wire to a rectifier 10 (e.g. transformer rectifier). The rectifier 10 is also connected to one or more anodes 18 which are also buried in the soil 12. The rectifier drives a current into the pipeline 14 through the soil 12.

[0042] Figures 2A and 2B illustrate a system for cathodic protection monitoring according to an example embodiment of the present application. As shown in Figures 2A and 2B, the example corrosion protection monitoring system includes one or more rectifiers 10 and one or more test stations 20. The example monitoring system 200 of the present application includes one or more remote monitoring units (RMU) for collecting data, the RMUs have a communications subsystem (e.g. transmit and possibly receive functionality) to allow communication over a network (e.g. cellular, satellite) with a monitoring platform 60. The monitoring platform 60 (e.g. CorView data platform) may be for example an external device, server, cloud server, or SaaS. There may be one or more types of RMU in the corrosion protection monitoring system. For example, RMUs may include rectifier monitors (e.g. impressed current driver measurement) and test stations (e.g. observers of potential). First system components may include one or more rectifiers, test stations, coupons, bonds, electrical and/or soil resistance probes, or pipe-to-soil test stations. Second system components may include one or more rectifiers, test stations, coupons, bonds, electrical and/or soil resistance probes, or pipe-to-soil test stations. In some instances, first system components and second system components may overlap and may be identical (e.g. both referring to rectifiers).

[0043] As shown in Figures 2A and 2b, each of the rectifier 10 includes one or more remote monitoring units (“RMU3”) 32 that may be mounted to or otherwise installed on the rectifier 10 or rectifier housing, for periodically measuring electrical features and transmitting data. In other embodiments the RMU3 32 may be positioned elsewhere.

[0044] In some embodiments, data is collected in one jurisdiction or country and is transmitted to another jurisdiction or country where data processing is performed. For example, the one or more RMU3 32 may be located in the United States and the monitoring platform 60 (e.g. server) may be located in Canada.

[0045] The rectifier 10 provide a near constant voltage output, with output controlled by manually configuring taps to set the desired voltage. As will be apparent to those skilled in the art, the rectifier could also be configured as a constant current device. The RMU3 32 is configured to take periodic readings of the electrical features of the system, to ensure the pipe is sufficiently protected from corrosion. These readings are then transmitted by satellite or cell network and are stored in a secure database. The RMU 32 may record the DC voltage, the DC current and the resistance of the cathodic protection system. The electrical current in a rectifier system travels through the soil to complete the circuit.

[0046] Rectifier monitors (e.g. RMUs, RMU3) measure the rectifier output voltage and current periodically at fixed intervals (e.g. every 6hrs). A secondary periodic transmission interval (e.g. one day to seven days) may also exist where the data is sent to a monitoring platform (e.g. CorView data platform, external device, server, cloud server) over satellite or cellular networks. Other communication forms and networks would also be possible, such as WiFi, internet etc.

The rectifier monitors are programmed with low and high thresholds for each of the measurement channels (e.g. voltage and current). If one or more of the measurements transitions in or out of alarm based on the thresholds, the RMU3 32 sends an exception transmission to the data monitoring platform immediately. In this way, if the corrosion protection system goes out of tolerance, it is not necessary to wait for the next periodic transmission interval to convey the system change.

[0047] The rectifier remote monitors are also capable of controlling the attached impressed current system in an on/off manner to facilitate the level of protection and close interval survey operations. The impressed current rectifier operational levels are typically set manually on the rectifier itself with no control by the remote monitor.

[0048] In other embodiments, the rectifier RMU3 32 may allow control of the rectifier operational parameters, thereby allowing remote adjustment based on feedback signals from the data platform (e.g. CorView). In addition, that feedback could take the form of an annual variation function that is transferred. That annual variation function may then allow local edge processing to determine the appropriate rectifier output levels or allow for changes in the measurement channel low/high thresholds. The former would optimize impressed current levels, while the latter would reduce nuisance alarms at the source.

[0049] Test station point monitors observe the effects of cathodic protection from impressed current and sacrificial anode protections. For example, the test station point monitors may include direct pipe potential observations and coupon observations. As shown in Figures 2A and 2B, for direct pipe observation of the cathodic protection current one or more test station remote monitoring units (“RMU1 Lite”) 36 may be used to sample the potential of the pipe with respect to a reference cell. Typically, “on” potentials will be captured at the measurement interval (6hrs). Again, measurements are sent on a periodic interval schedule (e.g. typically 14 or 28days). Channel high and low limits are also present with exception transmissions similar to the rectifier monitors.

[0050] In some embodiments, if the impressed current or sacrificial anode system protecting the pipe is put into survey mode (e.g. interruption mode), where it toggles the cathodic protection current on and off, the RMU1 Lite 36 is able to capture the instant-off potential measurements for the pipe. It does this by using a digital signal processing algorithm after capturing a one minute segment (at the 6hr measurement interval) of the pipe potential and then applying an autocorrelation algorithm to determine if a periodic waveform is present. If deemed periodic, the data is then run through a maxima/minima detection algorithm with outlier rejection to determine the on potential and off potentials. Additional qualifiers are present to avoid false detection.

Also, averaging of data is used to reduce effects of transients. The result of the interruption waveform measurements is sent to the monitoring platform (e.g. CorView data platform) immediately when a transition to/from interruption mode is detected.

[0051] Coupon monitors (“RMU1”) operate by taking measurements on a piece of metal with a known exposed face size made of the same material as the pipe; this is known as a CP coupon. Typically, on potentials, instant disconnect potentials and current flows of the coupon will be captured at the measurement interval (e.g. 6hrs). Again, measurements are sent on a periodic interval schedule (e.g. typically lhr to 28days). Channel high and low limits are also present with exception transmissions similar to the rectifier monitors. A benefit of the coupon monitor is that it allows equivalent instant-off potentials to be taken without having to interrupt the impressed current system that is protecting the pipe.

[0052] As shown in Figures 2A and 2B, the monitoring system 200 obtains data from various remote monitoring units (e.g. data collection units) which have two-way communication functionality. The data is then sent from the various remote monitoring units to the monitoring platform via a communication network (e.g. satellite network, cellular network etc.). In the example shown, rectifier monitor RMU3 32 is positioned at the rectifier 10 and adjusts the output levels of the rectifier. The rectifier transmits analog data in the form of changing pipe potentials to one or more test stations, and as shown in the figure test station 22 and a second test station 24. As well in the example, test station monitor RMU1 Lite 36 is positioned (e.g. mounted to or otherwise installed) at the first test station 22 and collects ground to soil readings as well as stores data received from RMU3 32. Also, test station monitor RMU1 34 is positioned (e.g. mounted to or otherwise installed) at the second test station 24 and collects coupon readings as well as stores data received from the RMU3 32. Data from the test station monitors (e.g. RMU1 34, RMU1 Lite36) and data from the rectifier monitor (e.g. RMU3 32) is sent to the monitoring platform 60 via a communications network (e.g. satellite, cellular). The data may be stored, managed and analyzed at the monitoring platform 60, including using machine learning algorithms to manage and optimize the performance of pipelines and cathodic protection systems. Data and control signals (e.g. based on data analysis and algorithm results etc.) may then be fed back to the one or more rectifiers 10 (e.g. feedback loop). For example, the rectifier set points may be controlled and adjusted remotely.

[0053] With the RMU3 32, seasonal variation in the rectifier’s voltage and current readings may be observed. This variation may be attributed to seasonal changes in soil resistivity. Figure 3 A illustrates a graph showing a weather seasonal variation in DC current of the cathodic protection system.

[0054] To better protect pipeline assets from corrosion, and to ensure that alerts do not become overwhelming, seasonally adjusted alarm limits are determined and provided. In the example method of the present application, this involves two phases. A first phase of predicted thresholds and a second phase of rectifier specific thresholds.

[0055] The following describes how seasonal variation of rectifier system resistance may be determined in accordance with an example embodiment. Resistivity of the soil in the impressed current cathodic protection (ICCP) rectifier system is not constant. Throughout the year, changes in temperature, precipitation and evaporation all play a role in changing soil resistivity. For a rectifier with fixed voltage output, the current will be inversely proportional to the resistivity of the rectifier-anode system, calculated using Ohm’s law. The system includes many components, but the fluctuations which have been observed may be primarily attributed to seasonal changes of soil resistivity. The voltage divided by the current may be expressed as the system resistance, which fits well to a sine, cosine or other periodic function, with a fixed period of one year. In a preferred embodiment: y = A cos(2 x — f) + y₀ (1)

[0056] As will be apparent to those skilled in the art, other periodic functions can be used to reasonably fit data of this type. [0057] In Equation 1, A is the amplitude of the annual variation in resistance, f represents the date shift of the maximum resistance, and y₀ is the average resistance. A cosine function was chosen instead of a sine function for simplicity of interpretation of the offset, the parameter f in a cosine function will indicate the maximum amplitude, or the calendar date with the highest resistance. An example of the cosine fit is shown in Figure 3B. In other embodiments other waveforms or functions may be used.

[0058] Figure 3B illustrates a cosine function fit of the seasonal variation of the DC system resistance (voltage/current). The amplitude and average resistance are in units of Ohms, and the phase offset is expressed in radians, with for example 0 representing January 1st and 2p representing December 31st. In this example, a phase offset of 1.04 would correspond to March 4th.

[0059] In order to predict thresholds, training data sets are used by a processing unit (e.g. monitoring platform, server, cloud server) as a baseline to predict future readings for a new rectifier location. The training data is used to generate classification and resistance fit parameters such as that shown in Tables 1 and 2. The resistance fit parameters are a statistical description of the thousands of example RMU readings contained in the training data. The dataset may be updated as more rectifier resistance readings are obtained and classified. Also, in future iterations of the machine learning model, the training dataset may be updated with additional rectifier readings. As well, the training data may include labelled or unlabelled datasets.

[0060] The use of machine learning provides a workable way to bring together and process or analyze disparate datasets, for example rectifier resistance data, geographic information, and environmental classification information, and other relevant data to generate datasets identifying rectifier fit parameters (and other parameters) that correspond to each category in a classification (e.g. soil temperature, soil moisture), as compared to non-ML implementations. It is not a process that can quickly, efficiently or effectively be performed by non-ML implementations. As such, the machine learning algorithm provides significant computing and processing capabilities that facilitate the generation of statistics corresponding to rectifier system resistance fit parameters (e.g. which could be stored in a database for subsequent use in the method and system of the invention) and include without limitation soil classifications.

[0061] The algorithms and processes in the present application may be implemented using a variety of programming languages with machine learning packages or modules such as the Python programming language as well as other programming languages such as Matlab, R, and Javascript. In Python, various packages may be used for modelling, such as scikit-learn package, TensorFlow, SparkML, H20, PyTorch. In the example embodiments, the algorithm of the present application is implemented in Python, using the scikit-learn package.

[0062] In an example embodiment for developing training data, the process for processing the seasonal (e.g. cyclical) data to cosine parameters, to broader statistics is provided. First, time-series measurements of rectifier resistance was obtained from over 3500 rectifiers across the United States. As well, data from impressed current cathodic protection (ICPP) rectifiers considered for this work has been collected for more than 12 years.

[0063] For each RMU3 32 specific resistance versus time dataset is cleaned of outliers (e.g. using the Inter Quartile Range method) and then the remaining data is fit with a cosine function, with the period of the function fixed to a certain time period (e.g. 1 year (365.24 days)). The remaining three parameters of the cosine function (e.g. amplitude, phase-offset, y-offset; see Equation 1) may vary.

[0064] Next, a least-squares fitting procedure was applied to the time-series measurements for each rectifier (e.g. find the best fitting curve). Figure 4 illustrates a histogram of phase offset fit parameters (e.g. other graphs, charts etc. may be generated to represent the data) according to an example embodiment of the present application. The phase offset represents the time of year with maximal resistance in the rectifier system. From the histogram in Figure 4, it is shown that the time of year with the highest system resistivity (e.g. peak in the histogram) occurs with a distribution centred near the middle of March. [0065] The sum of least squares approach to curve fitting may be implemented (e.g. using Python). The optimizer searches for the combination of the three parameters which best describes the data, by attempting to minimize the sum of the squared difference between the fitted curve and each real resistance reading. An example a well-fit set of readings is shown in Figure 11. In other embodiments, other curve fitting procedures may be used.

[0066] For every RMU3 32 with sufficient data (e.g. > 1 year of readings, > 30 individual readings (e.g. points in time data); approximately 10,000 RMU locations) the fit procedure was performed. Generally sufficient data means enough readings to unambiguously fit a reading set to a cosine function. The three optimized parameters (as well as a ‘goodness of fit’ parameter - R²) were stored in a table. In some embodiments, the fits which were not successful (R² < 0.5) were dropped from the table, as these units could not be described by a simple cosine curve. In other embodiments these datasets are fit by adding a slope, offset (constant) or higher order functions to the fitted curve.

[0067] Next, the rectifier resistance data may be compared to soil classification data from external data sources, such as for example the SSURGO database. Geographic regions in the US are classified based on measured soil features. Two classifiers were collected from the SSURGO database which reflect the broadest grouping (e.g. soil moisture and soil temperature). The SSURGO classifications are the current standard however other classifications and standards may be utilized.

[0068] The SSURGO (Soil SURvey GeOgraphic) Database contains information collected by the National Cooperative Soil Survey and hosted by the United States Department of Agriculture. Most counties in the continental US have data available (e.g. as shown in Figure 9).

[0069] Each county is further subdivided into Map Units. Map Units describe regions of unique soil properties, interpretations and agricultural productivity. These map units appear spatially as polygons of various sizes and shapes. Figure 10 shows an example of the various Map Units dividing a mixed use industrial/residential location. Each Map Unit contains features such as soil type, soil temperature class, soil moisture class, slopes, salt content, clay content, soil erodibility, soil conductivity and more. However, all features are not always present for every Map Unit.

[0070] In the example embodiment, the SSURGO database may be queried using SQL. For example, a Python script may be used to query the SSURGO database for each Map Unit containing a RMU3 32, finding the associated Map Unit for each RMU latitude/longitude. This generated a table of soil features with one entry for each RMU location.

[0071] Soil temperature classifications may be based on the mean annual soil temperature, the mean summer temperature and the difference between winter and summer temperatures at 50 cm depth. These range, for example, from Hyperthermic (hottest) to Pergelic (coldest). In the example embodiment of the present application, the coldest region discussed is Frigid.

[0072] Soil moisture classifications may be based on the typical levels of groundwater tables and amounts of soil water available to plants at a particular time of year. These range, for example, from Aridic (driest) to Aquic (wettest).

[0073] In an example embodiment, environmental data information (e.g. dataset) is collected and obtained for use in determining seasonal alarm thresholds. For example, for each rectifier its geographical position (e.g. latitude and longitude) is determined, and a query is made to the SSURGO database and the soil temperature and moisture regime classification may be obtained for that rectifier. Rectifier locations missing either soil temperature classification or soil moisture classification were dropped from this study. Approximately 1,100 rectifier datasets with soil classification were used.

[0074] RMU locations were labelled with their associated Soil Temperature and Soil Moisture classifications, the amplitude and phase-offset parameters for each classification were plotted for various Temp/Moisture labels (e.g. Figures 5, 6). Fit parameters and soil classifications plotting the fitted amplitude and phase offset of the cosine fit procedure for the temperature classification (Figure 5) and for the moisture classification (Figure 6) reveals that the various classes show very similar general fit trends. [0075] Grouped statistics were generated (Tables 1, 2) for cosine fit parameters for each temperature and moisture classification regime (e.g. represent resistance fit parameters for each temperature classification and for each soil classification). Table 1 illustrates a rectifier system resistance fit parameters (amplitude and offset) for various temperature classifications. Table 2 illustrates a rectifier system resistance fit parameters (amplitude and offset) for various moisture classifications. Table 1 and Table 2 present the accompanying statistics of the soil temperature classifications and the soil moisture classifications, where there one can observe separability of the classifications. [0076]

Table 1

Soil temperature classifications and rectifier system resistance fit parameter statistics

Table 2 Soil moisture classifications and rectifier system resistance fit parameter statistics

[0077] Temperature Amplitude : In Table 1, the amplitude (A) (e.g. median column) increases with classification from a warmer to a colder regime. This may be a result of the warmer classifications will have less change between winter and summer, and do not typically go through any freezing in the winter.

[0078] Temperature Phase offset. In Table 1, the phase offset (f) (e.g. median column) decreases for colder regimes. This reflects the resistance of the system reaching a maximum earlier in the year. This may be due the increase of resistance due to frost. This has been also observed by experimental measurement of soil resistivity and rectifier current above and below buried pipelines.

[0079] Moisture Phase offset: In Table 2, the phase offset (f) (e.g. median column) decreases for wetter regimes, again with resistance maximum shifting earlier into the year. This may be due to the increase of resistance due to frost.

[0080] Previous studies have highlighted that of the many variables in soil throughout the year, moisture plays the biggest role in modifying resistivity. However, in the case of buried pipelines and anode ground-beds, the seasonal variation of temperature may have a greater influence on the system resistance than seasonal moisture variation. Seasonal variation of soil moisture is greatly supressed for soil depths greater than one or two metres, whereas the seasonal temperature variation penetrates deeper into the soil.¹⁰ Additionally, a depth dependent temperature lag has been observed, meaning that coldest day of the year at the surface will occur days to months before the coldest day of the year at depth.¹¹ The observed maximum system resistance occurring mid- March may be attributed to the temperature lag at the depth of the pipeline-soil-anode system. As well, temperature and moisture classifications are not completely independent, with some correlation existing between soil temperature classifications and soil moisture classifications. [0081] There is the potential separability of the fit parameters by soil classification, and the potential dependence of fit parameters on soil temperature and moisture classifications (e.g. especially for the phase offset). Using the collected rectifier data, a decision tree model is provided that predicts a particular moisture classification or temperature classification by the fit parameters. That is, for a rectifier with sufficient data, the rectifier’s fit parameters (e.g. amplitude, phase etc.) may be used to determine the appropriate moisture classification or temperature classification, and thereby provide precise alarm thresholds for the rectifier (e.g. specific high and low resistance level limits). The classification logic may be implemented in other ways.

[0082] In the example embodiment, a decision tree was created (e.g. using the scikit-learn Decision Tree package in Python) to illustrate the classification logic for classifying a rectifier. The number of levels in the example decision tree is 3 which was determined to be optimal number for the present data, however the levels in the decision tree may vary for the same data or for different data or data combinations.

[0083] Decision trees need to have enough levels to allow for accuracy, without overfitting. Too few levels would result in a bad prediction of moisture and temperature regimes, and too many levels would result in overfitting, meaning the model would work well for the data used in training the model, but is not able to fit new data. In the example embodiment, three levels was decided upon by splitting the dataset into a training group and testing group, and attempting to maximize accuracy for both sets with various levels.

[0084] In other embodiments, the algorithm may be implemented by way of other models besides decision trees. For example, other machine learning approaches and supervised learning algorithms may be the basis for creating the decision algorithms, such as Neural Networks, Gradient Boosting Machines (GBM), Naive Bayes Classifiers, Stacked Ensembles, and XGBoost. Also, the machine learning algorithm may be supervised learning, semi-supervised or unsupervised. [0085] The decision tree splits the fit parameter data (e.g. amplitude, phase) by a specific variable at each level, determining the features of each of the splits (nodes) which result in the most accurate final classification. For example, the resistance fit parameters for a rectifier are input into the decision tree and based on the inputted values, the appropriate classification (e.g. soil temperature, soil moisture) for the rectifier is determined. By knowing the appropriate soil classifications for the rectifier, specific rectifier alarm thresholds are provided that will be more accurate for the specific environmental conditions the rectifier is located.

[0086] Figure 7 illustrates a decision tree diagram for soil temperature regime classification, according to an example embodiment of the present application. This decision tree resulted in an accuracy of 60%, meaning that by following the nodes of the tree and deciding to take the right or left branch based on a specified fit parameter the correct temperature classification (e.g. for the rectifier 10 location) would be achieved 60% of the time. The moisture classification tree’s accuracy was 55%, due to lower separability of moisture regime data (e.g. less variance of values between regions).

[0087] In the example decision tree of Figure 7, the first determination is based on an amplitude value. If the rectifier being evaluated has an amplitude value higher than a specified amount, then the classification process moves to the next level and evaluates phase offset value. If the rectifier amplitude is lower than the specified amount, the process moves to the next level and evaluates the constant(yo) value. The process continues until the rectifier has been classified into a temperature region based on its rectifier resistance parameters. By knowing the appropriate classification for the rectifier, alarm thresholds are provided that will be more accurate for the specific environmental conditions the rectifier is located. The decision tree may be ordered differently and may have different parameters evaluated at each node. For example, the ordering and parameters are determined by running thousands or more simulations and finding the parameters allowing the best prediction of the test data.

[0088] In one approach 80% of the available data is used for a training set, and 20% is used for a testing set. The parameters for the nodes of the decision tree are determined using the training set, and tested using the testing set. Notably, 100% accuracy is not sought, otherwise the decision-tree model will be overfit and the model is more likely to be fragile against future or real-world data.

[0089] To increase decision tree algorithmic accuracy in the future, a larger dataset of rectifier readings may be used, or additional features specific to the rectifier location or asset features may be considered. In particular, locally collected soil resistivity measurements, for example using a portable instrument and the Wenner method, at time of install may augment the accuracy of the algorithm.

[0090] Using the resulting data described above (e.g. soil temp/moisture classifications and rectifier system resistance fit parameter statistics) a two-phase approach may be used to suggest rectifier alert thresholds.

[0091] The system and method of the present application is driven by the machine learning algorithm which provides a framework for determining a soil classification (e.g. temperature, moisture) for a rectifier location and based on the classification, determine alarm threshold limits (e.g. upper and lower resistance limits) for the rectifier. These determined alarm thresholds are communicated to the remote monitoring unit (RMU) for the rectifier.

[0092] For predicted thresholds, upon installation of a new RMU3 32, the machine learning algorithm, which has been trained with historical rectifier readings, suggests an alert threshold. The suggested range may be based on publicly available soil and temperature classifications specific to the rectifier’s location, as well as features describing the pipeline and cathodic protection system. In other embodiments, other data and data sources may be utilized.

[0093] A process 1200 for generating predicted thresholds is shown in Figure 12, according to an example embodiment. Upon installation of a RMU3 32 on a rectifier 10 which has not previously been monitored, the data obtained and analyzed for soil classifiers may be taken to predict the seasonal variation in rectifier settings and to suggest alarm limits. First, information about the RMU3’s 32 geographical position is obtained (e.g. latitude and longitude) (block 1201). Next, environmental data is obtained for the rectifier’s geographical location (block 1202). The environmental data may be sourced from existing databases of historical environmental data (e.g. SSOGRO) or may be obtained from other sources. The environmental data may include for example soil temperature data and soil moisture data. Next, the soil classifiers (e.g. for soil temperature and soil moisture) are obtained (block 1203). Then, predicted threshold limits for alarms (e.g. upper and lower limits for voltage and current variation) are generated for the rectifier based on the soil classifiers (e.g. temperature and moisture) and the corresponding resistance parameter statistics for those classifiers (block 1204).

[0094] Rectifier specific thresholds may be implemented in the weeks and months after an RMU3 32 is activated. For example, after the collection of location specific readings for a sufficiently long time-period (e.g. dependent on frequency of measurement and transmission, may be greater than 1 year), an attempt can be made to perform a fit to a sine function of the rectifiers’ readings. If the fit is successful, location specific alert parameters can be determined based on the actual statistics of the readings on the rectifier. This allows rectifier readings (e.g. voltage, current) to be captured and a rectifier specific dataset is constructed. With the thresholds based on the data, the alert threshold becomes more focused and specific to the rectifier being monitored (e.g. specific to the environmental factors at the rectifier’s specific location). This approach will be more accurate than the other prediction method above, as location-specific factors will inevitably be accounted for with readings, rather than attempting to describe the average behaviour. The collection of location specific readings for a sufficiently long time- period (e.g. dependent on frequency of measurement and transmission, may be greater than 1 year), an attempt can be made to perform a fit to a sine function of the rectifiers’ readings. If the fit is successful, location specific alert parameters can be determined based on the actual statistics of the readings on the rectifier. This approach will be more accurate than the other prediction method above, as location-specific factors will inevitably be accounted for with readings, rather than attempting to describe the average behaviour.

[0095] A process 1300 for generating rectifier specific thresholds is shown in Figure 13, according to an example embodiment. First, the RMU3 32 collects location specific readings (e.g. voltage and current) over time (block 1301). After the collection of location specific readings for a sufficiently long time-period (e.g. dependent on frequency of measurement and transmission), an attempt may be made to perform a fit to a sine function of the rectifiers’ readings (block 1302). If the fit is successful (e.g. the data is fit to the curve) the parameter data (e.g. amplitude, phase offset) for the rectifier is obtained (block 1303). The proposed algorithm for generating rectifier specific thresholds is based on the cosine fit parameters obtained from the historic rectifier readings (e.g. voltage and current). These parameters, in addition to a factor based on the standard deviation (s) of the readings would define a range large enough to avoid unwanted alerts, yet still be focused enough to ensure the rectifier is operating as intended. The rectifier parameter data is inputted to evaluated it based on the soil classification logic (e.g. decision trees for soil moisture, soil temperature classification) to determine classifiers for the rectifier (block 1304). Location specific alert thresholds (e.g. upper and lower resistance limits) may be determined (block 1305). In this way, specific alarm thresholds are provided based on the actual statistics of the readings on the rectifier 10. This approach will be more accurate than the location threshold prediction method discussed above, as site specific variables will inevitably be accounted for with readings (e.g. provides rectifier specific thresholds).

[0096] Figure 8 shows an example plot of the implementation of these seasonally adjusted alert thresholds. The thresholds (e.g. limits) define a range of acceptable voltage and current readings (e.g. V/I).

[0097] In some embodiments, a level of protection survey may be performed at the minima and maxima points on the rectifier readings to establish adequate protection of the pipeline against standard protection criteria through seasonal changes. This may be accomplished through a manual survey if the pipeline is not equipped with test station monitoring. Alternatively, if the pipeline is equipped with remote monitored coupon test stations, or test station monitors that can detect interrupted sources and establish instant off measurements, the level of protection can be established with ease. This would give confidence that the seasonal alert thresholds are valid.

[0098] Effectiveness of cathodic protection of a pipeline is gauged by measuring the electropotential of the pipe along its entirety. Basic chemistry sets the minimum potential where the corrosion redox reaction will not occur. For steel pipelines, two primary potential criteria are used, the -850mV (relative to a CuCuS04 reference) and lOOmV criteria. In the case of the - 850mV criteria, the polarized potential of the pipeline is taken just after removing the drive from impressed current systems. This is known as the instant off potential — if this value is more negative than -850m V, then the pipe is deemed to be protected. The second criteria involves comparing the instant off potential to the depolarized potential. Here, the pipe is allowed to depolarize by removing the impressed current drives for an extended period of time and then the potential is measured. If the instant off potential is more negative than the depolarized potential by lOOmV, the pipe is also deemed to be protected against corrosion.

[0099] These measurements can be obtained in level of protection surveys where the measurements are taken at test stations that are spaced along the pipe for a check in a coarse granular manner. They can also be obtained from close interval surveys where the entire length of the pipe is walked, taking measurements every few feet. The close interval surveys have finer granularity, giving better visibility into susceptibility of coating defects that may be between test stations.

[00100] The combination of the level of protection and close interval surveys establish the protection for a given set of impressed current operational parameters and soil conditions. Based on these, the system could infer the protection levels throughout the year if measured at the maximum and minimum operating points. As well, with more frequent and granular readings from RMU1 34 on coupons and RMU1 Lite 36 on test stations, in other embodiments of the invention the monitoring system may be automatically adjusted (e.g. via the same algorithm developed for the seasonal variation of limits) to ensure optimal energy output (e.g. energy savings) and avoidance of over protecting (e.g. leading to hydrogen embrittlement). [00101] The use of rectifier DC voltage and current readings by RMU3s 32 throughout the

United States has allowed for a better understanding of the seasonal variation of rectifier system resistance. The soil component of the rectifier system electrical circuit is the biggest factor in seasonal variation. [00102] By comparing soil moisture and temperature regimes to this resistance data, it has been demonstrated that seasonal variation of resistance can predict the moisture and temperature classification of a rectifier’s location.

[00103] A two-stage procedure for prediction of future RMU alert thresholds has been proposed. The first stage involves a prediction of seasonally adjusted limits based on the rectifier’s location and the associated soil temperature and soil moisture classifications. The second stage will be rectifier specific seasonally adjusted limited based primarily on the previously recorded rectifier DC voltage and current readings.

[00104] Future analysis on the influence of asset depth, material and coating, as well as soil resistivity measurements at the time of install, may allow for a more accurate prediction of seasonally varying rectifier parameters.

[00105] Certain adaptations and modifications of the described embodiments can be made. Therefore, the above-discussed embodiments are considered to be illustrative and not restrictive.

Claims

1. A method for managing a cathodic protection system, the method comprising: obtaining data from one or more monitoring units; transmitting the data to a monitoring system via a communications network; analyzing the data using machine learning algorithms to manage one or more first system components in the cathodic protection system; and transmitting control signals to one or more second system components based on the data analysis and algorithm results.

2. A method for managing a cathodic protection system, the method comprising: obtaining data from one or more monitoring units; transmitting the data to a monitoring system; determining, from the data, limit set points for one or more first system components; and updating the limit set points for one or more second system components.

3. The method of claim 2 wherein the one or more first system components is one or more of a rectifier, coupon, bond or pipe-to-soil test station.

4. The method of claim 2 wherein the one or more second system components is one or more of a rectifier, coupon, bond or pipe-to-soil test station.

5. The method of claim 2 wherein the limit set points is an indication that a system component is disconnected or inoperable.

6. The method of claim 2, further comprising: obtaining data from one or more coupon test stations and from one or more pipe-to-soil test stations.

7. The method of claim 2 or claim 3, wherein the data is time-series voltage data or time- series current data.

8. The method of any one of claims 2 to 4, wherein the set points include voltage, current and resistance.

9. A method for determining the optimal DC output of a rectifier, the method comprising: obtaining data from one or more rectifiers, from one or more coupon test stations, and from one or more pipe-to-soil test stations; transmitting the data to a monitoring system; determining from the data an optimal level of one or more operational parameters based on the data; and adjusting the rectifier current to achieve the optimal level of the one or more operational parameters.

10. The method of claim 6, wherein the data is time-series voltage data or time series current data.

11. The method of claim 6 or claim 7, wherein the optimal level of DC current is based on regulatory compliance criteria.

12. A method for determining optimal DC output of one or more rectifiers, the method comprising: obtaining data from the one or more rectifiers; obtaining AC interference data; transmitting the data and AC interference data to a monitoring system; determining from the rectifier data and the AC interference data an optimal level of DC current at the one or more rectifiers; adjusting the DC output of the rectifier based on the determined optimal DC current level.

13. The method of claim 9, wherein the AC interference data is obtained from one or more electromagnetic field (EMF) sensors or from AC load data from a power utility.

14. A method for generating a classification dataset, the method comprising: obtaining data for one or more rectifiers; fitting the data using a fitting procedure to obtain fit parameters; determining a location for the one or more rectifiers; obtaining environmental data for the one or more rectifier location; classifying the one or more rectifiers based on the environmental data; and generating the dataset comprising data for each of the one or more rectifiers identifying classification and corresponding fit parameters.

15. A method for predicting alarm thresholds, the method comprising: determining a location of one or more rectifiers; obtaining environmental data for the location of the one or more rectifiers; classifying the one or more rectifiers based on the environmental data; and generating a predicted alarm threshold based on data statistics that correspond to the classification.

16. A method for determining alarm thresholds, the method comprising: obtaining data at one or more rectifiers over a period of time; fitting the data with a fitting procedure to obtain fit parameters; classifying the one or more rectifiers based on the fit parameters; and generating alarm thresholds based on the determined classification.

17. The method of any one of claims 11 to 13 wherein the data is time-series resistance data, time-series voltage data or time-series current data.

18. The method of claim 13, wherein the step of classifying comprises inputting the fit parameters into a classification logic.

19. The method of claims 11 or 12, wherein the environmental data is soil temperature data and soil moisture data.

20. The method of any one of claim 11 to 16, wherein the fit parameters include at least one of amplitude, phase offset and y-offset.

21. The method of claim 11 or claim 13, wherein the fitting procedure is a cosine function or a sine function.

22. A cathodic protection monitoring system, the system comprising: one or more monitoring units for collecting data; a monitoring subsystem, for receiving the data over a communications network, the monitoring subsystem comprising a processor configured to: determine, from the data, limit set points for one or more first system components; and update the limit set points for one or more second system components.

23. The system of claim 22 wherein the one or more first system components is one or more of a rectifier, coupon, bond or pipe-to-soil test station.

24. The system of claim 22 wherein the one or more second system components is one or more of a rectifier, coupon, bond or pipe-to-soil test station.

25. The system of claim 22 wherein the limit set points is an indication that a system component is disconnected or inoperable.

26. A cathodic protection monitoring system, the system comprising: one or more monitoring units for collecting data; a monitoring subsystem, for receiving the data over a communications network, the monitoring subsystem comprising a processor configured to: analyze the data using machine learning algorithms to manage one or more first system components in a cathodic protection system; and transmit, over the communications network, control signals to one or more second system components based on the data analysis and algorithm results.

27. The system of claim 22 or claim 26, wherein the one or more monitoring units includes rectifier monitors and test station monitors.

28. A computer program product comprising a non-transitory computer readable medium having instructions stored thereon, which when executed by a processor, the processor performs the method of any one of claims 1 to 21.