Asphalt density and compaction testing: measure and accept the mat

In-place density, and the number that outlives the paving

In-place density is how tightly the compacted asphalt mat ends up packed, and it is the single number that decides how long the pavement lasts. It is reported as a percent of Gmm, the theoretical maximum density of that mix. A mat at 93 percent of Gmm has 7 percent air voids. A mat at 90 has 10, and that gap does not show the day you pave. It shows in the cracking and the raveling a few winters early.

Density acceptance is a different job from getting the density. Rolling the mat hot, in the right pattern, inside the temperature window is how you build the density, and the companion compaction-window guide covers that side. This guide is about measuring what you built and getting paid for it: the reference values, the gauges, the cores, where to test, and the pay schedule that turns a number into money.

The trap is treating the test as paperwork after the real work is done. The test is the real work. A mat that rolled out clean and reads 91.5 on the core is a mat that loses a pay penalty and a chunk of service life, and nobody on the crew sees it until the cores come back from the lab. The crew that understands the acceptance test builds to it, not around it.

Why does asphalt density matter?

Density matters because air voids are how a pavement ages, and density is the one thing on the jobsite that controls them. Compact the mat well and the air voids are low, isolated, and sealed, so water and air cannot move through the asphalt. Leave it under-compacted and the voids climb, interconnect, and open the mat to the two things that destroy it: water getting into the structure and oxygen aging the binder.

The mechanism is not subtle. Below about 92 percent of Gmm, roughly 8 percent air voids, the voids stop being isolated bubbles and link into channels. Water moves through those channels, strips the binder off the aggregate, and the surface ravels and cracks. The binder oxidizes faster because air reaches more of it. A mat two points low on density ages on a faster clock than the one beside it that hit the number.

The numbers behind it are real. The long-standing rule of thumb across the agency research is that each extra 1 percent of air voids costs roughly 10 percent of service life, so the difference between a mat left at 92 percent of Gmm and one compacted a point or two higher shows up as years of fatigue and moisture life. Density is the biggest durability lever the field controls, which is exactly why agencies put incentive and penalty money on it through percent-within-limits pay factors.

What is the target air voids for asphalt?

The common in-place target on a dense-graded mat is about 7 to 8 percent air voids, which is the same as 92 to 93 percent of Gmm. Air voids and density are two ways of saying one thing: 100 percent of Gmm minus the percent density is roughly the air-void content, so 93 percent of Gmm is 7 percent voids by definition.

Do not confuse in-place air voids with the design air voids. The mix is designed in the lab to a target around 4 percent air voids at the design gyration level, and that is the volumetric target for the mix, not the field. The mat leaves the paver looser than the lab plug, and you compact it down toward, but not all the way to, that 4 percent. The field acceptance number, 92 to 93 percent of Gmm, sits above the lab design voids on purpose, because you cannot reach lab density on the road and you would not want to.

Treat the 92 to 93 figure as the common target, not a law. The agency specification sets the acceptance band, and those bands vary. Many specs run a band somewhere in the range of roughly 92 to 96 percent of Gmm, with a pay target a point or two above the floor, but the controlling number is the one in the project documents. Pull it before the first truck, because the band you are built to is the band you get paid against.

What is percent of Gmm?

Percent of Gmm is the in-place density expressed against the mix's theoretical maximum density. It takes two numbers. Gmm is the theoretical maximum specific gravity, the Rice value, the density the mix would have at zero air voids, run in the lab on loose mix under AASHTO T209 or ASTM D2041. Gmb is the bulk specific gravity of the compacted mat, the density you actually got. Divide Gmb by Gmm, multiply by 100, and that is your percent of Gmm.

Gmm is the reference and it has to be right, because every density on the project is measured against it. Get the Rice value wrong, from a bad sample or a mix that drifted off the design gradation, and every percent of Gmm on the job is off by the same amount. Gmm moves when the mix moves, so it is run on the day's mix, not assumed from the design.

Gmb comes two ways, and they are not interchangeable. The lab measures it on a cut core by weighing the core in air and in water, the bulk specific gravity test. The gauge estimates it in the field without cutting anything. The core Gmb is the true bulk density. The gauge reading is an estimate of it that has to be tied back to cores to be trusted. When this guide says density, it means Gmb over Gmm, however you got the Gmb.

Gmm (Rice value)

Theoretical maximum specific gravity of the loose mix, the zero-void density and the 100 percent reference (AASHTO T209 / ASTM D2041)

Gmb

Bulk specific gravity of the compacted core or mat, the in-place density (AASHTO T166 / T331)

Percent of Gmm

In-place density as a percentage of Gmm; 100 minus it is roughly the air-void content

Percent of Gmm versus percent of a control density

There is more than one basis for percent compaction, and confusing them is how two honest people argue past each other about a passing mat. The dominant basis today is percent of Gmm, the theoretical maximum, the one this guide uses. It measures the mat against the absolute ceiling of zero air voids, so the number ties straight to air voids and durability.

The older and still-used basis is percent of a control or target density, where the 100 percent reference is a lab-compacted density or a control-strip density instead of the Rice value. On that basis a number near 96 to 98 percent of the control density can be the target, and it sounds higher than a percent-of-Gmm number for the same mat, because the reference is lower. The two scales are not the same, and you cannot compare a number across them without knowing the basis.

So the first question on any density spec is which basis it uses. A spec calling for 92 percent and a spec calling for 96 percent can be asking for the same compacted mat, if one is percent of Gmm and the other is percent of a control density. Read the basis before you celebrate or panic over a number. The agency specification names it, and the test method follows from it.

Field example: turning Gmb and Gmm into a density

Put numbers on it. A dense-graded surface mix runs a Gmm of 2.485 from the day's Rice test. The lab cuts a core from a random spot in the lot and measures its bulk specific gravity, Gmb, at 2.305. Divide 2.305 by 2.485 and multiply by 100, and the core reads 92.8 percent of Gmm, which is 7.2 percent air voids. That passes a 92 percent floor, with not much room above it.

Now run the gauge at the same spot before the core comes out. The nuclear gauge reads 2.260, which is 90.9 percent of Gmm on its own. That does not mean the mat is low. It means the gauge needs its offset. The core is 92.8 and the gauge read 90.9 at the same point, so the gauge runs about 1.9 percent low on this mix and needs that correction added to every reading until the mix or the gauge changes.

Watch what one wrong input does. If the Rice value had been logged at 2.510 instead of 2.485, the same core Gmb of 2.305 would read 91.8 percent of Gmm rather than 92.8, and a passing lot would look like a failing one over a number nobody re-checked. The Gmm is the input that quietly moves every result, which is why it is run on the day's mix and verified, not carried over from last week.

What is a nuclear density gauge?

A nuclear density gauge is a portable instrument that reads the density of the mat in a minute or two without cutting it, using a small radioactive source and a detector that counts how gamma radiation interacts with the asphalt. The denser the mat, the more the radiation scatters and the fewer counts reach the detector, and the gauge converts that count into a density. The field method is ASTM D2950 and AASHTO T310.

It runs two ways. In backscatter mode the source sits at the surface and reads the gamma that bounces back, which keeps the gauge fully on top of the mat and is the usual asphalt mode. In direct transmission the source drops down a hole into the lift and the reading runs through the material, which is more common on soils and base than on a thin surface course. For most HMA acceptance work the gauge reads in backscatter.

The gauge has two costs people forget. The first is the standard count, a daily reference reading taken on a reference block to confirm the gauge has not drifted, and a gauge run on a bad or skipped standard count is reading against a moving baseline. The second is the source itself. The radioactive source means licensing, a trained and badged operator, locked storage, leak testing, and transport rules, and that overhead is exactly why the non-nuclear gauges exist.

Non-nuclear density gauges

Non-nuclear density gauges read the mat with no radioactive source, which takes the licensing, storage, and transport burden off the crew. The common type, the PQI and similar, works on electrical impedance: it sends a field into the mat and reads how the asphalt's electrical properties change with density. Set it on the mat, take a reading, move on. No badge, no source, no leak test.

The catch is that they read relative more than absolute. A non-nuclear gauge is sensitive to surface moisture, to the aggregate gradation and source, and to the mat temperature, so a reading that is dead-on for one mix can carry an offset on the next. That makes it strong for tracking the rolling pattern in real time, where you care that density is climbing pass over pass, and weaker as a standalone acceptance number unless the spec allows it and you have correlated it hard to cores.

Treat the non-nuclear gauge the same way you treat the nuclear one for acceptance: a fast field instrument that has to be tied to cores, not a referee on its own. Where it shines is control. The operator watches the number rise as the breakdown roller works, knows when the mat has stopped gaining, and saves the cores for the acceptance points. Wet the surface and the reading wanders, so dry it or wait.

Cores: the referee method

The core is the referee, and on most jobs it is also the acceptance test. You cut a full-depth plug out of the compacted mat with a core drill, take it to the lab, and measure its bulk specific gravity, the Gmb, against the Gmm. It is the most accurate in-place density you can get, because you are weighing the actual compacted material, not estimating it through a gauge.

The lab Gmb runs two ways depending on the mix. The saturated-surface-dry method, AASHTO T166 or ASTM D2726, weighs the core in air and submerged and is the standard for dense mixes that do not drain. For coarse or open mixes where water runs into the surface voids and throws the SSD weight off, the vacuum-sealing method, AASHTO T331, seals the core in a bag and gives a truer number. Sampling the loose mix and handling the cores follow the established practices, with ASTM D979 covering sampling of the mixture.

Cores cost you. They are slow, they leave holes you have to fill, and you do not get the answer until the lab runs it, often after the crew has moved a mile down the road. That is the price of the truth. The gauge gives you speed during rolling and the core gives you the number that settles acceptance and a dispute, and when the two disagree, the core wins every time.

Why does the gauge have to be correlated to cores?

A density gauge reads an estimate, not the truth, and the estimate carries an offset that depends on the mix, so the gauge has to be corrected to cores before its numbers mean anything. The gauge does not know your aggregate, your gradation, or your binder. It was calibrated on reference material, and the difference between that reference and your actual mix is the offset you have to find.

You find it by taking gauge readings at spots where you also cut cores, comparing the two, and computing the average difference. That offset gets added to or subtracted from every gauge reading on that mix until something changes. Change the mix design, the aggregate source, or even the gauge, and you redo the correlation. A common error is correlating once at the start of a job and trusting it after the mix has drifted, which leaves you tracking a number that quietly stopped matching the cores.

This is why a gauge number alone never settles acceptance. The gauge reading of 90.9 against a core of 92.8 in the field example is not a low mat. It is a gauge that needs its 1.9 point correction. Run the gauge for control during rolling, correlate it to cores, and let the cores carry the acceptance. A gauge accepted on a bad correlation is a number that looks like data and behaves like a guess.

Where and how often to test

Where you test is set by the spec, and it is built around randomness and lots. The pavement is divided into lots, often by tonnage placed or by area or by a day's production, and each lot gets a set number of tests at random locations chosen so the crew cannot steer the gauge to the good spots. Random stationing and random offsets across the width are the point. A test you got to pick is not an acceptance test.

Stay off the very edge. The unconfined outside edge of a pass and the area within a foot or so of any free edge read low, because the mat had nothing to compact against there, so acceptance tests are taken in from the edge by the distance the spec sets. The number of tests per lot and the lot size both come from the agency specification and the level of acceptance, statistical or not, that the spec runs on.

The frequency is not arbitrary. A spec running statistical acceptance needs enough tests per lot to compute a meaningful estimate of the lot's quality, which is why those specs call for several random cores or readings per lot rather than one. Take the tests the spec asks for, at the random spots it picks, in from the edge, and the lot result means something. Cut corners on any of those and you have a number that does not represent the mat.

Longitudinal joint density

The longitudinal joint is the lowest-density spot on the mat, and it gets its own test and often its own pay line because everyone knows it runs lean. The joint is the seam between two adjacent paving passes, and the unconfined edge of the first pass had nothing to compact against, so it ends up several points below the field of the mat. That lean seam is exactly where water gets in and the pavement starts to ravel and crack, years before the rest of it shows age.

Because the joint is the known weak point, many agencies set a separate minimum joint density and tie pay to it. The minimum commonly lands somewhere around 90 percent of Gmm by the SSD core method, a couple of points below the field target, with the exact number and the pay adjustment varying by agency and test method. A well-built joint runs only 1 to 2 percent below the surrounding mat. A bad one runs 5 to 10 percent low, and that is the joint that opens in two winters.

Test the joint where the spec says, which is a core or a gauge reading centered on or just off the seam, not out in the field of the mat where the number looks better. How you build joint density, the edge restraint and the rolling, is covered in the compaction-window guide. The acceptance point here is simpler: the joint is measured separately because it fails separately, and the spec puts money on it for that reason.

Density lives inside the compaction window

You only get density while the mat is hot enough to move under the roller, so every acceptance number traces back to whether the mat was rolled inside its temperature window. Density is built while the binder is fluid and the aggregate can slide into a tighter packing. Once the mat cools below its cessation temperature the skeleton sets, and no number of passes closes another void. A cold mat reads low on the core and there is no fixing it after the fact.

This is why a density failure is usually a temperature or timing failure wearing a different shirt. The mat that came back at 90.5 was often laid thin, on a cold base, in wind, or with a roller train that fell behind and gave the heat away before the breakdown pass was done. The acceptance test catches the result. The compaction window is where the cause lives.

The companion compaction-window guide covers the temperatures, the roller train, the test strip, and the cooling clock in depth. The link to acceptance is direct: plan the rolling to the window, hit the pattern that the test strip proved reaches density, and the cores confirm what you built. Skip that and the gauge tells you in real time that the mat stopped gaining before it reached the number.

Segregation and density

A segregated mat will not compact to density no matter how you roll it, because the problem is in the material before the roller ever touches it. Segregation is the separation of the mix, either the coarse aggregate concentrating in spots, called aggregate segregation, or cold streaks of mix, called thermal segregation. Both leave patches that are coarser or colder than the surrounding mat, and both read low on density.

You see aggregate segregation as a rocky, open-textured patch, often in streaks down the center or at the edges where the material dropped out of the paver. You see thermal segregation with a thermal camera or an infrared bar behind the screed, as cold spots that came off the truck or the windrow colder than the rest. A cold or coarse spot has more air voids the day it is laid, and a density test that happens to land on one reads the segregation, not the rolling.

The fix is upstream of compaction, in the material handling: even truck exchange, a material transfer vehicle to remix and even out the temperature, and a steady head of mix in front of the screed. When a density result comes back low and the spot looks rocky or the infrared showed a cold streak, the answer is not more rolling. It is finding why the mix separated before it hit the mat.

Thin lifts and density

A lift placed too thin will not reach density, for two reasons that stack. The first is heat. A thin mat has little mass and a lot of surface, so it sheds heat fast and the compaction window slams shut before the rollers finish, leaving voids locked in. The second is the stone. The lift has to be thick enough relative to the largest aggregate that the stones can shift and seat instead of bridging on each other.

That second one is the t over NMAS rule, the ratio of lift thickness to nominal maximum aggregate size. The common minimum is roughly 3 times the NMAS for fine mixes and about 4 times for coarse mixes, so a 1/2 in stone mix wants a lift on the order of 1.5 to 2 in, not 3/4. Go thinner and the mat cannot densify, it turns permeable, and it is prone to segregation on top of everything else.

The acceptance consequence is direct. A scratch course spread too thin over a rough surface is a mat that was never going to make density, and chasing it with the roller only marks it. Match the lift to the stone and to the heat it needs to hold, and the density target becomes reachable. The compaction-window guide carries the t over NMAS detail. The point for acceptance is that a too-thin lift fails the test by design.

How is asphalt density accepted and paid?

Density acceptance runs from a simple pass-fail against a minimum up to a statistical system that sets your pay by how consistently the lot hit the target, and the modern agency trend is the statistical one. The most common statistical method is percent within limits, PWL, which samples the lot, looks at both the average and the variability, and estimates what percentage of the lot actually falls inside the specification limits.

PWL pays for consistency, not just for clearing a floor. A lot that runs tight and centered on the target earns a high PWL and can earn incentive pay above the contract price. A lot that swings around, even if the average passes, earns a lower PWL because more of it sits outside the limits, and that draws a pay penalty. Specs commonly set an acceptable quality level near 90 PWL for full pay and a floor, often around 75 PWL, below which the lot does not qualify for payment at all, with the numbers varying by agency.

The reason the field cares is money on the table. Density is one of the pay factors with real incentive and penalty attached, because the agency research ties density straight to service life. Hit the target consistently and the PWL can pay a bonus. Run loose and you give it back. The exact limits, the lot size, the number of tests, and the pay schedule all live in the project specification, so build to that spec, not to the rule of thumb.

Intelligent compaction and continuous coverage

Intelligent compaction puts instruments on the roller so the crew can see compaction across the whole mat in real time instead of at a handful of test points. An IC roller carries an accelerometer on the drum, GPS positioning, an onboard display, and a temperature sensor, and it builds a color-coded map of pass count, mat temperature, and a stiffness-related value the FHWA calls the ICMV across 100 percent of the area, not the small fraction a gauge or a core samples.

What it changes is coverage and uniformity. A core tells you the density at one spot. The IC map tells you where the roller missed a pass, where a cold streak ran, and where the mat stiffened unevenly, so the operator can fix coverage before the mat cools instead of finding a soft spot weeks later in a core. It catches the temperature segregation and the missed passes that produce the low cores in the first place.

The technology is past the pilot stage. The FHWA pushed it through its innovation programs and a couple dozen states have written asphalt IC specifications, though the ICMV does not yet replace cores for acceptance on most jobs. Treat it the way the better specs do: a process-control and coverage tool that drives the density up and the variability down, with cores and gauges still carrying the acceptance number.

Why is my asphalt density too low?

When a core or a gauge comes back below target, the cause is almost always one of a short list, and they rank by how often they bite. The mat was rolled too cold, which is the most common: the breakdown pass came late, or the lift was thin, or the base and the wind pulled the heat before the pattern was done. Temperature and timing are the first place to look every time.

Next is the roller pattern itself. Not enough coverages, the wrong roller for the mix, a vibratory drum set wrong for a thin lift, or no pneumatic in the train to chase the voids a steel drum leaves. Then the mix: a tender mix that shoved instead of compacting, or a segregated spot the test happened to land on. Then the support, a soft or unstable base under the mat that moves under the roller and eats the compactive effort instead of letting the mat densify against it.

The fix depends on which one it is, and the compaction-window guide carries the rolling side in depth. The acceptance discipline is to find the cause before the next load, not after. A correlated gauge running during rolling tells you the mat stopped gaining before it reached the number, which is the warning the cores cannot give you until it is too late to change anything on that lot.

When air voids are too low

Density can be too high, and that failure is as real as the low one, just less famous. Push the air voids too far down, below roughly 3 percent in service, and the mat has no room for the binder to move into on a hot day or under traffic densification, so it flushes and shoves. The binder migrates to the surface, the mat gets slick, and it ruts under load. Both ends of the air-void range are failures, which is why specs run a band, not just a floor.

A too-low result usually points at the mix more than the rolling. A rich mix with too much binder, or a mix with the wrong gradation, compacts down past the safe void range and then bleeds. You can also over-compact a normal mix in a hot climate by piling on far more passes than the test strip called for, though it is harder to do than people think. The more common cause of low voids on a finished pavement is a mix that left the plant out of spec.

The point for acceptance is that the spec band has a top as well as a bottom, and a number near the high-density end deserves a look at the mix, not a victory lap. If cores come back at 97 or 98 percent of Gmm with the binder flushing, the conversation moves from the roller to the plant. Hitting the target means landing inside the band, not maximizing the number.

Heavy-duty and data-center pavements

On heavy-duty pavements, the ones carrying concentrated loads or built to last decades, density acceptance gets tighter because the cost of getting it wrong is higher. Industrial yards, port and intermodal pavements, freight-heavy routes, and the access and yard paving around large facilities like data centers and distribution centers all see loads and a service-life expectation that punish a few extra air voids hard.

The physics does not change, but the margin does. The same interconnected voids that shorten an ordinary road's life shorten a heavy-duty pavement's life faster, because the loads are heavier and more frequent and the water that gets into a lean joint does more damage under that traffic. So the specifications on these jobs tend to run a tighter density band, more tests per lot, and stricter joint requirements, and the owner often has its own spec on top of the DOT method.

The lesson for the crew is that the density discipline that earns a bonus on a highway is the price of entry on a heavy-duty job. Tight rolling, correlated gauges, honest joint density, and a clean record are not optional where the pavement has to carry real load for thirty years. Build to the stricter band, document it, and the pavement does the job the owner paid for.

What to document

The density record is what defends a marginal lot when the cores come back light, and it is what gets a crew paid for a borderline number instead of penalized for it. A density result with nothing behind it is a number nobody can reconcile or dispute. The crew that wrote down the Gmm, the gauge offset, the test locations, and the temperatures is the crew whose lot survives the pay meeting.

Capture the mix and lift, the Gmm from the day's Rice test and who ran it, the test method and which gauge, the gauge-to-core offset and when it was last correlated, the test locations and how they were chosen, each density result as a percent of Gmm, the joint density separately, the lot definition and the resulting pay factor or PWL, and any retest or dispute core. If a lot ran low, note the temperatures and the conditions that explain it, because that note made at the time is worth more than a memory at the dispute.

Common mistakes

• Accepting a gauge number that was never correlated to cores, or correlated once and trusted after the mix drifted.
• Running the nuclear gauge on a bad or skipped standard count, so every reading rides a moving baseline.
• Testing the cold, unconfined edge instead of taking the acceptance test in from the edge as the spec requires.
• Ignoring the longitudinal joint, which is measured and paid separately because it fails first.
• Carrying over yesterday's Gmm instead of running the Rice value on the day's mix, so every percent is off by the drift.
• Spreading a lift thinner than the t over NMAS minimum, so it cannot reach density and the roller only marks it.
• Testing a spot that landed on a segregated or cold streak and blaming the rolling for material that separated upstream.
• Treating density as paperwork after the pour instead of building the rolling to the acceptance test.
• Maximizing density past the spec band into flushing and rutting instead of landing inside it.

Field checklist

Standards and references

The density numbers rest on a stack of AASHTO and ASTM test methods, and naming the right one for each step is half of defending a result. Theoretical maximum specific gravity, Gmm or the Rice value, is AASHTO T209 and ASTM D2041. Bulk specific gravity of the compacted core, the Gmb, is AASHTO T166 or ASTM D2726 by the saturated-surface-dry method, with AASHTO T331 vacuum sealing for coarse or open mixes that drain. Sampling the mixture follows ASTM D979.

The in-place field tests have their own methods. Nuclear density measurement is ASTM D2950 and AASHTO T310, in backscatter or direct transmission, against a daily standard count and correlated to cores. Non-nuclear electromagnetic gauges have their own procedures and the same correlation requirement. The Asphalt Institute's MS-22, Construction of Quality Asphalt Pavements, is the practical trade reference for placement, compaction, and density acceptance.

What governs a given job is the agency specification on top of these methods. The state DOT or owner spec sets the density basis, the acceptance band, the lot size, the number of tests, the joint requirement, the statistical method, and the pay schedule, and it varies by agency. The targets in this guide, the 92 to 93 percent of Gmm, the joint near 90, and the PWL figures, are the commonly cited numbers. The project documents and the adopted specification set the numbers you build and get paid to, so confirm them before you rely on any single value.

Units, terms, and conversions

The same compacted mat gets described in a few names and unit systems, so a density reads differently across a Rice report, a core lab sheet, a gauge readout, and a spec.

Density is reported as a percent of Gmm, and 100 minus that percent is roughly the air-void content, so 93 percent of Gmm is 7 percent air voids. Specific gravities, Gmm and Gmb, are unitless ratios to the density of water and run to three decimals, where a thousandth of a point matters. Temperature is in °F on most U.S. paving, density bands and air voids are in percent, and lift thickness is in inches against an aggregate size in inches or millimeters, so a 1/2 in NMAS is 12.5 mm. Percent within limits, PWL, is a percentage of the lot estimated to fall inside the spec limits, not a density itself.

Gmm (Rice value)

Theoretical maximum specific gravity of the loose mix, the zero-void density and the 100 percent reference (AASHTO T209 / ASTM D2041)

Gmb

Bulk specific gravity of the compacted core or mat, the in-place density (AASHTO T166 / T331)

Percent of Gmm

In-place density as a percentage of Gmm; 100 minus it is roughly the air-void content

Air voids

The percentage of the compacted mat that is air, the inverse of percent of Gmm

PWL

Percent within limits, the share of a lot estimated to fall inside the specification limits, used for statistical acceptance and pay

Standard count

A daily reference reading that confirms a nuclear gauge has not drifted before it is used

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• What temperature do you stop rolling asphalt?
• What is the pavement condition index (PCI)?
• What is intelligent compaction?
• Why is asphalt density important?
• Why do asphalt joints fail before the rest of the pavement?