Increasing Predictability and Investor Confidence in PV Power Plants through Latent Defect Screening

Alex C. Mayer and Jenya Meydbray

PV Evolution Labs, Berkeley, CA, 94710, USA

Abstract – Solar power plant investors expect photovoltaic (PV) modules to safely and efficiently produce electricity for 25 years. International certification standards such as IEC are designed to evaluate modules for defects and design flaws that contribute to product safety or early lifetime performance issues. This certification testing is performed on a small number of pre-production panels. The majority of solar panel issues observed in the field, however, are driven by deviations in the manufacturing process, not by fundamental design flaws. These off-specification manufacturing defects are typically latent. This means that the panels initially meet performance expectations, but suffer accelerated performance degradation. Accurate data on the percentage of panels that exhibit significant latent defects is hard to come-by, but several studies suggest that the industry rate is around 4%. The appearance of latent defects significantly increases operating costs for the installation. The ability to gain knowledge of the exact quality of the PV panels installed at a power plant provides opportunity for improved output predictability and investor confidence. This knowledge will be increasingly important as the market penetration of PV increases, especially considering the more than 600 module suppliers. There is currently no certification to insure against PV panel underperformance caused by latent defects. In this article we introduce the concept of latent defect screening for PV modules. Latent defect screening involves the random sampling and accelerated-life testing of the PV panels to be installed at the construction site. We find that for an additional system cost of 1 penny per watt, we can be 95% sure that there are fewer than 3% defects in a 20MW installation.

Index Terms — Reliability, certification, project finance.

I. INTRODUCTION

An accurate forecast of a PV power plant’s output is

paramount to lowering the levelized cost of electricity (LCOE)

and obtaining favorable financing terms. In today’s market,

the lack of output predictability leads to increased system and

operating costs. This inaccuracy is mostly due to the fact that

only 5% of installed panels have been in the field for 10 or

more years. [1] As more information becomes available, the

system costs will be lowered due to increased investor

confidence, reduced cost of capital, and lower insurance

premiums. [2] At the same time, higher accuracy reduces the

need for a large reserve on the grid and PV plant curtailment

to handle PV output variability. [3, 4]

A large deviation from the annual forecasted electricity output

appears when a fraction of the PV modules exhibit latent

defects. A latent defect occurs when a panel initially meets

performance expectations, but manifests a defect that causes

accelerated performance degradation or can lead to a safety

issue such as an electric shock or an electrical fire. Appendix 1

outlines the various types of latent defects seen in the field.

Defects such as solder-joint and junction-box degradation

(Figure 1) are not necessarily caused by a design flaw, but

rather by deviations in the manufacturing process that lead to

compromised product quality.

Fig. 1: Photographs of typical latent defects. Solder joint

degradation (a) and junction-box arcing (b).

It is a common misconception that Underwriters laboratory

(UL) and International Electrotechnical Commission (IEC)

certification ensure product quality. [5 - 7] However, this

testing is only performed on a handful of pre-production

prototypes and mostly aims at ascertaining the quality of the

materials and product design. Certification does not ensure

manufacturing quality control; any deviation due to material

supply, tool aging, process drift, etc. can lead to failures in the

field.

Most panel manufacturers and system owners are,

unfortunately, hesitant to share experiences regarding actual

field failure rates. Actual experience ranges from 0.1% to 10%

and occasionally up to 100% (Table 1). [8 - 11] These failure

rates are expected to increase as the more than 600

978-1-4673-0066-7/12/$26.00 ©2011 IEEE 001643978-1-4673-0066-7/12/$26.00 ©2011 IEEE 001643

manufacturers face a downward cost pressure that incentivizes

manufacturers to cut corners.

TABLE I: REPORTED PV PANEL FAILURE RATES

These latent defects lead to lost revenue in several ways:

• Reduced power production; the defective panel produces

less electricity for the time between the defect’s initial

occurrence and the panel replacement. In some cases, the

response can take several months to verify the defect and

execute the manufacturer‘s warranty. Given the series-

connected nature of PV arrays, the loss of power can be

substantial, as a reduction in performance of one panel

will affect the entire string.

• Defective panel replacement costs; these costs include

logistics, labor, and powering down a string of modules to

make the replacement.

• Increased operation and maintenance costs associated

with panel inspections to find other defective units; panel

defects caused by manufacturing deviations typically

occur in clusters. In other words, a manufacturing facility

may produce many consecutive good panels followed by

several consecutive defective panels. This increases the

likelihood of multiple defective panels at an installation

since the panels used for a project are generally

manufactured around the same time.

Each project owner will have to compute the exact estimate

for the replacement costs per Watt (based on geography,

module type, module supplier contracts, labor costs, etc.).

Actual replacement costs can vary between $0.30 and

$3/Watt. [11, 12] Using a reasonable estimate of $0.50/W, we

can make a ballpark estimate for added cost to the project

owner. This translates to an additional system cost of $0.02/W

for 4% defects. [9] On a 100MW installation, this can turn into

an extra two million dollars – not including the added cost of

the required conventional reserve, insurance, interest rates,

lost energy, and damage to a company’s reputation.

TABLE II: CERTIFICATION TESTING PROTOCOLS

The best way to avoid financial penalties associated with

latent defects is through advanced screening. Latent defect

screening consists of random sampling and accelerated life

testing to ensure panel quality for a given installation. Table 2

summarizes the differences between standard design

certifications and latent defect screening. This accelerated

testing destroys the panel, takes 20 to 60 days, and costs

money, thus making it prohibitively expensive to test every

panel before deployment. As we will now show, larger

sample-sizes ensure that the panels are of acceptable quality.

Nomenclature

N Number of panels for an installation

n Sample-size

C Number of defects found in sample

fmax Max percent defective

α Confidence-level associated with fmax

$repl Per panel replacement cost

$risk Potential cost associated with partial

panel replacement

$test Per panel testing cost

For the last 50 years producers and customers have negotiated

acceptable quality level (AQL) sampling plans to assure the

quality of manufactured products. [13] These plans provide a

guide for the number of units to randomly test to guarantee

with high confidence that the percentage of defective units is

less than a max percent defective, fmax. These AQL plans

suffer from a large sample-size requirement to assure quality,

which increases testing costs. Large sample-sizes are

necessary to allow for several defects to be detected through

testing.

Designed to Evaluate

Timing Sample Size

Certification Design Prototypes 8 – 12 panels

Latent Defect

Screening

Manufacturing

StabilityPer Project

Statistically

Significant

WhenVolume Affected

What Occurred

2008420,000

modulesManufacturing defect

2008-2009 ~$215M Loss of performance

2002-2008300,000

modules

Deteriorating

insulation

200554,000

modules

W eak cell

interconnects

1994-20020.13% return

rateVarious failures

early 2000’s >10% Junction box fires

early 2000’s ~3.5% Severe cell cracks

early 2000’s 2.90%

Solder joint failure

causing localized

heating

978-1-4673-0066-7/12/$26.00 ©2011 IEEE 001644978-1-4673-0066-7/12/$26.00 ©2011 IEEE 001644

By limiting the number of defects in the sample, C, to 0 or 1,

the sample-size can be greatly reduced. [14] In this case, if a

defect is found through testing, the consumer can either reject

the incoming product or require further testing on more units.

With this requirement, the relationship between the

population-size, N, the testing sample-size, n, the confidence-

level, α, and the maximum percent defective, fmax, can be

calculated using the hypergeometric probability function. The

probability that 0 defects are encountered in testing is given

by:

. (1)

The parameter α – sometimes called the “producer’s risk” –

represents the “confidence-level” that there are less than fmax

defects in the population. By increasing n, a lower fmax is

generated at a given confidence. If one defect is encountered,

the numerator of equation1 becomes ����������� �������

� � .

Approximations are given in reference 15.

The trade-off between the sample-size and product quality

assurance is shown in Figure 2 for a 20 MW installation at the

75% and 95% confidence-levels. The calculations based on

the hypergeometric distribution are shown for the cases where

0 and 1 latent defects are encountered during testing. For the

case where every element of the sample passes the screening

(solid and broken lines in Figure 2), the assured max percent

defective reduces rapidly with sample-size. Finding 0 defects

out of a sample-size of 74 panels ensures, with95%

confidence, that less than 4% of the panels in the 20 MW are

defective. If the sample-size is doubled to 148, we can be 95%

confident that less than 2% of the panels will exhibit a latent

defect in its lifetime. If on the other hand, the customer and

lender are content with 75% confidence, then these sample-

sizes ensure a max percent defective of 0.9% and 1.9% for

sample-sizes of 148 and 74, respectively. If the supplier is a

trusted name, this confidence-level may be acceptable to the

project financier for use in their calculations.

Fig. 2: Effect of sample-size on the max percent defective for

a 20 MW installation.

If a sample-size of 74 were chosen – expecting to ensure less

than 4% defects – and one defect was exposed, we would be

95% confident that there were less than 6.2% defects for the

20 MW installation (outlined circles, Figure 2). For this

situation, the customer could demand further testing or send

the panels back to the manufacturer. In the case of further

testing, we would have to measure another 42 panels – finding

no more defects – to be 95% confident that there were fewer

than 4% defects for a 20 MW installation. Depending on

testing costs, this large sample-size may be acceptable.

TABLE III: CONFIDENCE-LEVEL AND MAX PERCENT

DEFECTIVE FOR A 20 MW INSTALLATION FOR

DIFFERENT SAMPLE-SIZES.

The financial benefit of testing can be shown by comparing

the cost of testing to the financial risk of finding latent defects.

For simplification, we will assume a fixed cost of testing, $test,

of $2,000 per panel. This number can vary depending on the

sample-size and testing details. A simple method to estimate

financial risk per Watt, $risk, is to multiply the replacement

cost per Watt, $repl, by the max percent defective and the

probability that there are more than fmax defects. The last term

is equal to one minus the confidence level. This gives:

( )

⋅

−

===

n

N

Nf

n

fN

P0

1

-1)fn,|0(c

maxmax

max α

fmax, fmax,

C = 0 C = 1

75% 0.90% 1.80% 148

75% 1.90% 2.60% 74

95% 2% 3.20% 148

95% 4% 6.20% 74

Confidence-Level

Sample-size

978-1-4673-0066-7/12/$26.00 ©2011 IEEE 001645978-1-4673-0066-7/12/$26.00 ©2011 IEEE 001645

$���� = $���� ∙ ���� ∙ �1 − �. (2)

The comparison between the cost of testing and the financial

risk for a 20 MW installation at 95% confidence is shown in

Figure 3. For this case, if a sample-size of 100 is chosen, the

additional testing cost is less than 1₵/W. This sample-size

corresponds to the 95% confidence-level for an fmax of 2.9%.

Table 4 lists the confidence-level and fmax for a 1 and a 20

MW installation at a fixed testing cost of 1₵/W. The financial

risk versus testing sample-size is calculated using a

replacement cost of $0.5/W. As can be seen, by testing ~40

panels the financial risk and the testing costs are

approximately the same.

Fig. 3: Financial risk and testing costs for a 20MW installation

as a function of sample-size.

TABLE IV: CONFIDENCE AND FMAX AT A FIXED

TESTING COST OF 1₵/W

While many panel producers have experienced substantial

defect rates there are, of course, many manufacturers who

have produced millions of panels with an exposed latent defect

rate of less than 4%. For these manufacturers, it becomes

important to understand how a complete picture of their panel

quality history changes the confidence associated with testing.

As more information becomes available, we can use this

knowledge to reduce sample-sizes (Appendix 2).

In conclusion, latent defect screening can be used to assure

solar panel quality for a given installation. This screening does

not replace IEC and UL certifications that assure the quality of

product design, but rather acts to ensure an installation against

deviations in the manufacturing process. This new, product

quality assurance screening fosters investor confidence, which

can help reduce project soft-costs, like insurance premiums

and debt-servicing payments.

REFERENCES

[1] S. Price and R. Margolis, “2008 Solar Technologies

Market Report” (2010).

[2] B. Speer, M. Mendelsohn, and K. Cory, “Insuring Solar

Photovoltaics: Challenges and Possible Solutions”,

NREL Technical Report 6A2-46932 (2010).

[3] “Large Scale PV Integration Study”, NVEnergy Report

(2011).

[4] “Integrating Renewable Electricity on the Grid”, A

Report by the APS Panel on Public Affairs (2010).

[5] IEC 61215: “Crystalline silicon terrestrial photovoltaic

modules – Design qualification and type approval”; IEC

61646: “Thin-film terrestrial photovoltaic modules –

Design qualification and type approval”; IEC 71730:

“Photovoltaic module safety qualification, part 2:

Requirements for testing”.

[6] ANSI/UL 1703: “Safety standard for flat-plate

photovoltaic module and panels”.

[7] G. TamizhMani, “Testing the reliability and safety of

photovoltaic modules: failure rates and temperature

effects”, PV-Tech (2010).

[8] M. Kanellos, “REC to Recall All of Its Solar Panels

From 2008: Report”, GreenTechMedia (2009).

[9] D. DeGraaf, R. Lacerda, Z. Campeau, ”Degradation

Mechanisms in Si Module Technologies Observed in the

Field; Their Analysis and Statistics”, presented at NREL

2001 Photovoltaic Module Reliability Workshop (2011).

M

[10] M. Osborne, “Manufacturing cost per watt at First Solar

falls to US$0.76: module faults hit earnings” PVTech

(2010).

[11] “Customer Friendly”, PHOTON: The Photovoltaic

Magazine, p. 81, Issue 9 (2011).

[12] “Utilizing Panel-Level Monitoring to Improve Project

ROI”, Alternative Energy Magazine (2012)

http://www.altenergymag.com/emagazine/2012/01/utilizi

ng-panel-level-monitoring-to-improve-project-roi-/1836

[13] ANSI/ASQ Z1.4-2003: “Sampling Procedures and

Tables for Inspection by Attributes” (2003).

[14] N. Squeglia, “Zero Acceptance Number Sampling

Plans”, ASQ Quality Press, Milwaukee, WI (1994).

fmax,

C = 0

75% 1 MW 24% 5

95% 1 MW 56% 5

75% 20 MW 1.40% 100

95% 20 MW 2.90% 100

Confidence-

level, αProject Size

Sample-

size, n

978-1-4673-0066-7/12/$26.00 ©2011 IEEE 001646978-1-4673-0066-7/12/$26.00 ©2011 IEEE 001646

[15] Write out the terms in the combinations in eq. 1 and

recall that Nf choose 0 is 1;

,

Canceling out the n! and writing the terms, we come to:

.

Writing out the first term, we find that we get (N-n)⋅(N-

n-1)⋅⋅⋅(N-Np-n+1) and for the second we get 1 over

(N)⋅(N-1)⋅⋅⋅(N-Nf+1). Each product has Nf terms and we

can approximate each term by the mean-value. This

gives:

Now we can solve for n and get

� = �

��1 − �1 − ��/������2" − "���� + 1. A

similar calculation can be made for C = 1.

( )

( )( )

( )!!

!

!!

!

1

nNn

N

nNfNn

NfN

−⋅

−−⋅

−

=−α

( )

( )

( )

!

!

!

!1

N

NfN

nNfN

nN −⋅

−−

−=− α

( )

( )

Nf

NfN

nNfN

+−⋅

−+−⋅=−

125.0

2125.01 α

978-1-4673-0066-7/12/$26.00 ©2011 IEEE 001647978-1-4673-0066-7/12/$26.00 ©2011 IEEE 001647