Automated root cause analysis

The present invention relates in general to an automated root cause analysis. In particular, the present invention is di ^¬ rected to finding the root cause of a turbine like a turbine or wind turbine.

Finding the root cause for misbehaviour of a certain product like e.g. a turbine, can be very time consuming, and is often characterized as a "search in blindness" in a vast amount of data available on a given turbine, or a group of turbines. The data can include construction BOM, service events, pa ^¬ rameter settings, software version, profiles, operational data, etc.

It is very difficult to search for explanations and reasons in distributed databases, and the data connected to a certain turbine is maintained in different departments of a company. A construction bill of materials (BOM) can be located at a new unit and service alterations can be at a service depart ^¬ ment etc.

When a turbine or a group of turbines is identified to be out of normal behavior either by the turbine controller (alarm monitoring) , by service engineering or by model based monitoring (dynamic limits) investigation on this is generally performed by technicians on site.

Only in special cases, remote analyses are performed. This remote analysis involves creating a picture of what has hap ^¬ pened to this turbine since it was produced, to make sure that any event that may have caused the misbehavior is found. The data needed for this analysis is to be found in many dif ^¬ ferent databases, in different departments, and in different structures. This is very time consuming, and there are big chances that something is missed that could point to the root cause. It actually turns out in many cases that when the root cause is known, one can also find an explanation in the kept data, but it was hidden due to the amount and complexity of the available data.

It is therefore an object of the present invention to improve root cause analysis.

This object is solved by the features of claim 1. The depend- ent claims offer further details and advantages of the inven- tion . The invention is directed to an automated root cause analysis for a complex product, with the following steps:

- Monitoring the product;

- Detecting a product misbehaviour;

- Forming a misbehaviour pattern with timeslots of normal operation and timeslots of misbehaviour operation;

- Comparing a timeslot of the misbehaviour pattern to corresponding timeslots of channels of information in at least one database relating to the product;

- Measuring how close the misbehaviour pattern fits to the information; and

- Automatically identifying the best fit.

The automated method of root cause analysis can be performed specifically in a business where a high number of very com ^¬ plex products are controlled from one position and informa ^¬ tion on the product is highly distributed. This applies very well to wind turbine service and is also adaptable to trans ^¬ portation and other industries. Automating the root cause analysis will reduce the time to be spent on remote trouble ^¬ shooting, operational follow up and technical support. The method according to the invention allows very fast identification of serial damages and enables to decide an action up-front . The channels and/or the information can be aligned. This can include all available data from all turbines and make them available in a structured way like binning on timestamps, binning continuous measures etc. The information and/or the at least one database can relate to a group of products instead of solely to the certain prod ^¬ uct. This broadens the base for comparisons and enhances the chances of finding the root cause, for example in finding ex ^¬ planations in the turbine structure by using bills of materi- als or the like.

Several channels preferably two to five may be utilised for the fit. Depending on computational power, even more channels can be used. The usage of multiple channels improves the method of root cause analysis for distributed causes or er ^¬ rors, which may depend on several even independent condi ^¬ tions .

One database may contain a bill of materials from the produc- tion of the product which allows finding correlations even from before the turbine is constructed by using lists from sub suppliers as versions, batches, series, age of components or the like. A matrix may correlate one event of misbehaviour to a data ^¬ base and/or a channel for the comparison of timeslots. For the matrix a programming/mathematical calculation software like for example ADA can be used. The use of such a matrix reduces the amount of data that has to be compared thus eas- ing the whole procedure. A misbehaviour pattern may be detected before it leads to a failure of the product and a failure of the product may be predicted and prevented. A service or emergency operation flag can be set or actions can be scheduled immediately or at the next planned service.

An indicator indicating the fit of the misbehaviour pattern to the information and/or channel can be calculated. The in ^¬ dicator can be a number in the range from one to hundred wherein a number of hundred can indicate the best fit. This indicator eases the following evaluation and processing.

An ordered list of indicators can be generated. In this list the indicators can be ordered according to their number or the like allowing a prioritized representation of explana ^¬ tions for the root cause of the misbehaviour.

The product may be a turbine. The term turbine encompasses a wind turbine (complete or parts like the generator) or a tur- bine used e.g. for power generation.

The accompanying drawings are included to provide a further understanding of embodiments. Other embodiments and many of the intended advantages will be readily appreciated as they become better understood by reference to the following de ^¬ tailed description. The elements of the drawings do not nec ^¬ essarily scale to each other. Like reference numbers desig ^¬ nate corresponding similar parts. Fig. 1 illustrates a schematic flow chart of a root cause analysis according to the invention.

Fig. 2 illustrates a schematic view of a first example of a root cause analysis according to the invention.

Fig. 3 illustrates a schematic view of a second example of a root cause analysis according to the invention. Fig. 4 illustrates an example of a matrix correlating certain misbehaviour to a database according to the invention. In the following detailed description, reference is made to the accompanying drawings which form a part hereof and in which are shown by way of illustration specific embodiments in which the invention may be practised. In this regard, di ^¬ rectional terminology, such as "top" or "bottom" etc. is used with reference to the orientation of the Figure (s) being de ^¬ scribed. Because components of embodiments can be positioned in a number of different orientations, the directional termi ^¬ nology is used for purposes of illustration and is in no way limiting. It is to be understood that other embodiments may be utilized and structural or logical changes may be made without departing from the scope of the present invention. The following detailed description, therefore, is not to be taken in a limiting sense, and the scope of the present in ^¬ vention is defined by the appended claims.

Figure 1 shows a flow chart of the automated root cause anal ^¬ ysis. Although several steps are shown and described not all steps are mandatory for the invention. The first step for ex ^¬ ample can be omitted.

As a prerequisite for automated root cause analysis, all the available data can be aligned in a first step 1. This can be done not only for the single product to be monitored but for all products of the same or similar kind. Here, a turbine is taken as an example. Accordingly, data from all turbines is taken and made available in a structured way. This can in ^¬ clude binning on timestamps, binning continuous measures, etc. This process can be on going and is not necessarily a part of the invention. In a second step 2 the turbine is monitored. As the turbine is monitored to detect turbine misbehaviour technical parame ^¬ ters of the turbine are monitored. In a third step 3 detection for turbine misbehaviour takes place. In normal operation no misbehaviour is detected and the process flow goes back to step 2 for further monitoring. Once a turbine misbehaviour is detected the process or method branches to a fourth step 4. The misbehaviour can be an error or a value being out of a normal band. The turbine misbehav ^¬ iour is the input for the automated root cause analysis. The misbehaviour can be, for example a rise in alarm events, findings at scheduled services, service engineering and also a model based monitoring which will detect turbines with ab- normal behaviour below the fixed alarm limits which can be referred to as dynamic limits.

In step 4 a misbehaviour pattern 5 with timeslots of normal operation and timeslots of misbehaviour operation 6 is formed. The misbehaviour pattern 5 is shown in the top of

Figure 2. This turbine misbehaviour is clustered to form this pattern 5 that can be compared to information in all the da ^¬ tabases containing relevant information. In detail the proc ^¬ ess will identify timeslots of when the turbine was in normal operation and when not.

In a further step 7 one, several or all timeslots of the mis ^¬ behaviour pattern 5 are compared to channels 8 of information in at least one database relating to the turbine. It can be sufficient to automatically compare only the timeslot of mis ^¬ behaviour 6 to corresponding timeslots 9 of the channels 8.

The method receives its input manually or automatically as a cluster/group of turbines and with the time where this/these turbines were judged to be out of normal behaviour. This is then the misbehaviour pattern. The method searches for correlations from even before the turbine is constructed by lists from sub suppliers on ver ^¬ sions, batches, series, age of components etc. From the time of construction, databases can include a bill of materials with detailed information on all components in the turbine, version sizes, ratings, ages etc. From the commissioning of the turbine, databases can include information available only to a project before handover to service. Thereafter, the en ^¬ tire lifespan of the turbine can be accessed like software versions, retrofits, change of settings, adjustments, parts replaced, defect codes of the replaced parts etc. Including these databases gives the complete history of all turbines.

For example, by comparing data from the production bill of materials, the method can identify bad deliverances from sub suppliers like a batch with bad or wrong components.

The timeslots of misbehaviour 6 of the misbehaviour pattern 5 are compared to the corresponding channels 8 or timeslots in all other databases like shown in Figure 2.

The automatic scan of all databases available can be steered with a matrix 9 like an ADA matrix shown in Figure 4. According to this matrix 9 only certain databases or channels 8 are compared which match to the input source or the type of the misbehaviour .

In a next step 10 it is measured how close the misbehaviour pattern 5 fits to the information and/or the timeslots 9 in the channels 8. A function is provided to measure how closely the pattern fits and an indicator 11 or value or weight from one to one hundred is set which is shown in the right column of Figure 2. The indicator 11 is the result of a calculation to indicate the fit to a corresponding pattern or channel 8. These indi ^¬ cators 11 will then form a list 12 of the most likely expla- nations to the unwanted behaviour. The method may show many, few or no pattern fit.

An iteration process may initiate to compensate for time dif ^¬ ferences or time lags as shown in the lowest line in Figure 3 ( " zoom" ) .

In a next step 13 the best fit is automatically identified. The best fit between the misbehaviour pattern 5 or the time- slot of misbehaviour 6 and the channel 8 or the corresponding timeslot 9 can be chosen among the indicators 11 or from the list 12.

Additionally, the method can also cluster a group of turbines in order to find explanations in the turbine structure using for example the bills of materials.

Some human intervention may be required to reason upon the findings. A final report is reported automatically and is forwarded to the personnel responsible to correct the prob ^¬ lems and/or to further systems.

In Figure 3 a situation is shown where several conditions ex ^¬ ist that need to be met to identify the misbehaviour. When, for example a turbine with a specific retrofit needs a corre ^¬ sponding software or software version the method uses two channels 8 to make the correlation. The correlation corre ^¬ lates the two corresponding indicators which have a value or weight of 6% and 8%, respectively to a secondary indicator 14 having a value of 100%.

The method can handle a correlation of more channels, like for example two to five channels. The number of channels ma depend on computational power .

As an additional feature the system can find and recognize misbehaviour patterns that will lead to failure before a failure actually occurs. By that approach, turbine failure is predicted and prevented by taking preventive actions remotely or at a next planned service. This feature may rely on pre ^¬ cise feedback on service events.